A substation intelligent early warning method based on data analysis

Abstract

The application discloses a substation intelligent early warning method based on data analysis and relates to the technical field of power monitoring and intelligent early warning, and comprises the following steps: establishing a unified time anchor network, performing microsecond-level time alignment on electrical signals of each collection channel in the operation process of a substation, forming a continuous time scale band, and providing a time reference for subsequent signal rhythm arrangement; based on the continuous time scale band, rearranging the sampling rhythm, setting staggered sampling intervals between the collection channels, separating high-frequency signals in the time domain, constructing a staggered peak sampling spectrum, and avoiding high-frequency signal frequency overlap; the application constructs a full-link intelligent early warning mechanism, accurately identifies abnormal features hidden by aliasing in high-frequency disturbance, effectively reduces the false alarm and missed alarm rates, realizes spectrum stability, accurate identification, constant early warning threshold, and significantly improves the substation operation state perception and active defense capability.

Description

Technical Field

[0001] This invention relates to the field of power monitoring and intelligent early warning technology, specifically to an intelligent early warning method for substations based on data analysis. Background Technology

[0002] Data-driven intelligent early warning for substations refers to utilizing multi-source operational data generated during substation operation, such as voltage, current, frequency, power factor, temperature, vibration, and switch status, through continuous acquisition, dynamic comparison, and multi-dimensional feature analysis to identify potential abnormal trends and latent fault signs in real time. This process no longer relies on manual inspections or single threshold judgments, but rather discovers early risk signals through correlations and changes between data, achieving early detection and precise location of problems such as equipment overheating, insulation aging, abnormal loads, arc discharge, and harmonic instability. The system extracts operational characteristics in the analysis phase, determines risk levels in the identification phase, and implements adaptive protection through parameter correction, energy allocation, or control strategy optimization in the dynamic control phase. This forms a closed-loop early warning system comprised of data-driven mechanisms, state perception, intelligent decision-making, and automatic intervention, enabling substations to shift from "passive alarm" to "active defense."

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, intelligent early warning of substation operating status typically relies on frequency domain analysis and trend judgment of multi-source electrical signals. However, when the system is under high-frequency disturbance background, signals such as current, voltage, and power factor from different acquisition channels are prone to simultaneously entering the spectral aliasing region. Frequency components fold and mirror interference occur, causing true high-frequency components to be masked by false frequency components during feature extraction. This aliasing phenomenon causes the analysis model to incorrectly focus on false feature regions caused by interference in the subsequent identification stage, thus forming a seemingly stable but internally abnormal false early warning blind zone. This problem is particularly prominent in operating scenarios such as dynamic load switching and concentrated harmonic bursts, potentially leading to misjudgment of the system's operating status, delays in fault identification, and in severe cases, serious consequences such as protection logic lag, overheating of electrical equipment, or cascading tripping.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a data analysis-based intelligent early warning method for substations to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a substation intelligent early warning method based on data analysis, comprising the following steps:

[0008] Establish a unified time anchor network to perform microsecond-level time alignment of electrical signals from each acquisition channel during substation operation, forming a continuous time scale band to provide a time reference for subsequent signal rhythm arrangement;

[0009] The sampling rhythm is rearranged based on the continuous time scale band, and staggered sampling intervals are set between each acquisition channel to separate high-frequency signals in the time domain and construct staggered sampling spectrum to avoid frequency overlap of high-frequency signals.

[0010] Identifiable weak marker pulses are injected into key sampling nodes of the staggered sampling spectrum. The reflected path of the marker pulses in the frequency domain is used to draw the mirror trajectory and generate a mirror trajectory map to locate the aliasing region in the signal spectrum.

[0011] Based on the mirror trajectory map, an anti-phase suppression window is set in the aliasing contamination frequency band. By weakening the energy of the mirror component at the edge of the window, the real high-frequency signal channel is restored, and a list of real signal channels is formed.

[0012] The arrangement order of the feature signal channels is reconstructed based on the real signal channel list, and the real high-frequency signal segments are concentrated into a unified identification sequence to generate a set of early warning anchor points, which are used as the core input in the dynamic monitoring stage.

[0013] By performing time-reversal phase traction operations around the early warning anchor point set, a miniature phase curtain is laid in front of the early warning anchor point. Combined with an energy bypass structure and adjustable virtual impedance, phase inverse traction and energy migration are performed, thereby constructing a dynamic control closed loop with stable spectrum and constant early warning threshold.

[0014] Preferably, the continuous time scale band formation process is as follows:

[0015] A unified time reference is established. The reference clock signal is output by the high-stability temperature-compensated crystal oscillator in the main control device and transmitted to the time receiving unit of each acquisition channel via a bidirectional optical fiber link. The phase adjustment circuit is used to make the local clock signal and the reference clock signal phase consistent.

[0016] Perform periodic time deviation verification by calculating the trigger time deviation by sampling voltage, current, temperature, operating current and power factor signals, and injecting a synchronization pulse packet to correct the sampling timing when the deviation exceeds the limit;

[0017] Establish a characteristic time stamp sequence, embed equally spaced calibration pulses into the time axis of each acquisition channel, and verify the consistency of the time scale by matching the stamp intervals;

[0018] The continuous verification process is performed, the time drift rate is recorded, and the synchronization accuracy is maintained through timing pulse phase compensation, thereby forming a stable continuous time scale band.

[0019] The preferred method for constructing the staggered sampling spectrum is as follows:

[0020] Based on the continuous time scale, the sampling time of each acquisition channel is redistributed so that the sampling actions are staggered on the time axis and the sampling windows of each channel do not overlap.

[0021] The sampling rhythm is determined by combining the main frequency characteristics of the signals from each acquisition channel, and the sampling interval difference is set according to the frequency range of voltage signals, current signals, temperature signals and vibration signals, so that different signal types form a fixed interlaced structure on the time axis;

[0022] A staggered sampling spectrum is generated using a continuous time scale band, and the signal independence between the time domain and the frequency domain is verified by cross-correlation coefficient calculation.

[0023] Periodically perform sampling rhythm verification and drift compensation, and adjust the sampling trigger delay to maintain the stability of the staggered sampling spectrum and the continuity of the time structure.

[0024] Preferably, the steps for generating the mirror trajectory map are as follows:

[0025] In the staggered sampling spectrum, key sampling nodes are determined based on signal energy density and phase stability. Injection nodes are selected at positions where the instantaneous amplitude of the signal is more than twice the average amplitude and the phase change between adjacent sampling points does not exceed a fixed angle.

[0026] At the identified key sampling nodes, weak marker pulses with controlled amplitude are injected. The pulse width remains constant, the rising and falling edges have the same slope, and the pulse injection is synchronized with the sampling rhythm to form a recognizable pulse sequence.

[0027] A frequency domain transformation is performed on the sampled signal after the injection of the marker pulse, and the reflected mirror trajectory is plotted based on the phase change and energy reduction law of the pulse reflection peak.

[0028] A mirror trajectory map is formed based on the reflected trajectory, and the specific location and range of the aliasing region are determined by the spectral energy distribution.

[0029] The preferred method for forming the actual signal channel list is as follows:

[0030] Based on the energy distribution of the mirror trajectory diagram, the frequency range and time boundary position of the anti-phase suppression window are determined, the frequency band is divided into multiple equal-width sub-segments, and energy smooth transition zones are set on both sides of the window to prevent signal breakage.

[0031] In a defined frequency band, phase inversion suppression is performed by superimposing a phase inversion signal within the aliasing band to weaken the mirror energy peak and setting an energy protection band at the edge of the window to maintain spectral balance.

[0032] The high-frequency signal after phase inversion suppression is extracted, and the real channel is restored based on energy and phase selection to form a list of real signal channels with recorded channel frequency range, amplitude value, phase offset and time scale position.

[0033] Preferably, the process for generating the early warning anchor point set is as follows:

[0034] The reconstruction priority of the channels is determined based on the actual signal channel list, and the channels are sorted according to the average energy density and phase fluctuation amplitude, and the signal reconstruction order is determined.

[0035] The characteristic signal channels are arranged and reconstructed according to priority order. The signals of each channel are realigned on the time axis based on the time scale band, and a frequency cross-fusion band is set at the channel junction to maintain signal continuity.

[0036] A set of early warning anchor points is generated based on the reconstructed recognition sequence. Energy anchor points and phase anchor points are determined by detecting the rate of change of signal energy and the amplitude of phase change. An effective set of early warning anchor points is generated based on the time position difference between the two types of anchor points.

[0037] Preferably, in the process of generating the early warning anchor point set, the determination of energy anchor points and phase anchor points is based on the energy change rate and phase change amplitude of adjacent time periods. When the time position difference between the two types of anchor points is less than the preset time threshold, it is determined to be a valid early warning anchor point. The time position, frequency center, signal amplitude, energy gradient and phase offset of each early warning anchor point are recorded as parameters to form a complete anchor point set.

[0038] Preferably, a time-reversal phase-pull operation is performed around the early warning anchor point set. A miniature phase curtain is laid in front of it, and phase inverse pull and energy transfer are performed in combination with an energy bypass structure and adjustable virtual impedance. The dynamic control closed loop steps are as follows:

[0039] Execute the time reversal trigger operation, generate a backpropagation signal with the early warning anchor point set as a reference, and set a time reversal trigger window in front of the anchor point to form a local time reversal interval;

[0040] A miniature phase curtain is laid inside the time reversal window. The phase distribution and propagation speed of the reverse signal are controlled by a phase modulation unit, and impedance matching layers are set at both ends of the phase curtain to keep the signal propagating continuously.

[0041] An energy bypass structure is constructed behind the micro phase curtain, so that the inverted energy can migrate in layers along the conduction channel and dissipate through the energy dissipation element;

[0042] An adjustable virtual impedance is introduced into the energy bypass path, and a dynamic phase compensation structure is formed by combining inductive and capacitive elements.

[0043] Closed-loop verification is performed on the signal after phase pulling and energy transfer to evaluate spectral stability and the constancy of the warning threshold.

[0044] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0045] This invention constructs a full-link early warning mechanism that achieves source synchronization, process separation, feature calibration, and stable output by establishing a unified time anchor network, recompiling sampling rhythms, injecting identification markers, suppressing mirror aliasing, reconstructing identification sequences, and performing time-reversal phase traction operations. Compared with existing techniques that rely on fixed frequency domain features and single threshold judgments, this invention can accurately identify hidden high-frequency feature signals masked by aliasing in the context of multi-source high-frequency disturbances, detect abnormal signs in advance, and greatly reduce the probability of false alarms and missed alarms. In continuous dynamic monitoring, the system can maintain stable spectral peaks, clear identification paths, and constant early warning thresholds, effectively avoiding protection malfunctions, identification delays, and equipment damage caused by false spectral feature interference, and significantly improving the intelligent identification capability and proactive defense level of substation operating status. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0047] Figure 1 This is a flowchart of a data analysis-based intelligent early warning method for substations according to the present invention. Detailed Implementation

[0048] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0049] This invention provides, for example Figure 1 The substation intelligent early warning method based on data analysis, as shown, includes the following steps:

[0050] Establish a unified time anchor network to perform microsecond-level time alignment of electrical signals from each acquisition channel during substation operation, forming a continuous time scale band to provide a time reference for subsequent signal rhythm arrangement;

[0051] To ensure the time synchronization of multi-channel electrical signals during substation operation and the consistency of their spectrum during subsequent analysis, establishing a unified time anchor network is a fundamental step in the entire early warning method. This process is completed step by step through high-precision time synchronization, time deviation verification, time marker identification, and continuity verification, thereby forming a stable and traceable continuous time scale band, providing an accurate time reference for signal rhythm arrangement and dynamic analysis. The specific implementation steps are as follows:

[0052] A unified time reference is established across all acquisition channels in the substation. A high-stability temperature-compensated crystal oscillator within the main control unit is selected as the reference clock source, with a clock frequency of 10MHz and frequency stability controlled within ±0.5ppm. The reference clock signal is transmitted to the time receiving unit of each acquisition channel via a bidirectional fiber optic link, with a fiber optic transmission distance of 2500 meters. The signal transmission delay is accurately measured using round-trip ranging. After receiving the reference pulse, each acquisition channel uses a local phase adjustment circuit to ensure that the phase of its local clock signal is consistent with the reference clock, with a phase error of less than 0.3 microseconds. To ensure the continuity of clock synchronization, the time reference pulse sequence is retransmitted every 100 milliseconds to detect phase drift and perform real-time correction. Unlike traditional timestamp-based alignment methods, this method eliminates the uncertainty of network transmission delay through physical layer clock phase adjustment, enabling high reliability and repeatability of time synchronization for multi-channel signals.

[0053] After establishing initial time anchors, periodic time deviation checks are performed on all acquisition channels to ensure the continuity of the time scale. One second is defined as a complete check cycle. During this period, the electrical signals of each acquisition channel are sampled, including six parameters: three-phase voltage, three-phase current, equipment temperature, oil level temperature, circuit breaker operating current, and power factor. After sampling, the deviation between the trigger time of each signal and the reference clock is calculated. When the deviation exceeds 0.5 microseconds, an automatic correction procedure is triggered. The correction procedure injects a 100-microsecond synchronization pulse packet into the channel in the next time window of the check cycle. Each pulse packet contains five consecutive pulses with a 20-microsecond interval. The receiver detects the arrival time of the peak of each pulse and determines the current channel delay value by comparing the peak phase difference. The trigger time of the local sampling timing control unit is adjusted to advance or delay the trigger time by the corresponding delay amount, thereby restoring synchronization. Under 32-channel parallel acquisition conditions, this process can reduce the synchronization error from 2.8 microseconds to 0.2 microseconds on average, ensuring that all channels are continuously distributed on the same time scale.

[0054] After the time scale band is formed, a characteristic time marker sequence is established to clarify the scale structure and improve the recognition capability and robustness of the time axis. Each time marker sequence consists of 5 equally spaced calibration pulses with a pulse interval of 100 microseconds and a pulse width of 2 microseconds. This marker sequence is superimposed on the time axis of each acquisition channel at a fixed period, so that identifiable pulse markers appear on the time axis of each channel. After data acquisition, the consistency between the channel and the reference clock can be quickly confirmed by analyzing the interval and waveform characteristics of the marker pulses in each channel. If the voltage signal channel detects 50 complete marker sequences within a 100-millisecond time period, and the current signal channel detects the same number and interval of markers within the same time period, then the two channels are fully synchronized. If the pulse interval offset in the detection result exceeds 0.2 microseconds, instant calibration is performed. This method has higher anti-interference capability than using only a single timing pulse, and can maintain the continuity and recognizability of the time reference even under noise disturbance or signal reflection conditions, thus making the time anchor network traceable in the subsequent data fusion process.

[0055] After the time-stamped sequence is established, a continuous verification process is performed to verify the reliability and long-term stability of the time anchor network under dynamic operating conditions. A 5-second interval is defined as a verification interval, with two adjacent intervals used as comparisons. At the end of the first interval, the time drift rate of each acquisition channel is recorded, and the average drift rate is calculated. When the drift rate exceeds 0.05%, phase compensation is performed before the start of the second interval. Phase compensation is achieved by changing the phase angle of the timing pulse; the adjustment angle is calculated based on the drift time. For example, when the drift time is 2.4 microseconds, the phase angle is adjusted to 8.64 degrees. After phase compensation, the phase recovery effect is confirmed by sending 10 consecutive sets of calibration pulses. If the peak offset of the 10 consecutive sets is less than 0.1 microseconds, the calibration is considered successful. After 5 verification cycles, the average synchronization error of all channels stabilizes within 0.15 microseconds. This process ensures that the time scale maintains linear growth characteristics under long-term operating conditions and is not affected by temperature changes, electromagnetic interference, or signal reflections causing time base shifts. The resulting unified time anchor network has microsecond-level synchronization accuracy throughout the entire acquisition cycle, providing a reliable time basis for staggered sampling, mirror trajectory plotting, and spectrum reconstruction.

[0056] The sampling rhythm is rearranged based on the continuous time scale band, and staggered sampling intervals are set between each acquisition channel to separate high-frequency signals in the time domain and construct staggered sampling spectrum to avoid frequency overlap of high-frequency signals.

[0057] After establishing a unified time anchor network and obtaining a continuous time scale band, to avoid spectral aliasing of multi-channel high-frequency signals during acquisition, the sampling rhythm needs to be rearranged. Signal separation is achieved through time-domain staggered sampling, thereby constructing a staggered sampling spectrum. This step involves precisely defining the sampling interval, rearranging the trigger timing, and performing hierarchical verification of spectral independence and continuous stability checks to form a high-resolution and traceable sampling timing structure. The specific implementation steps are as follows:

[0058] Based on a continuous time scale, the sampling times of each acquisition channel are redistributed to create a strictly staggered distribution of sampling actions on the time axis. A sampling period of 1 millisecond is set, and this period is subdivided into 1000 sampling time slots, each 1 microsecond in length. In practical implementation, it is assumed that there are 32 acquisition channels, corresponding to voltage signal channels 1 to 8, current signal channels 9 to 16, temperature signal channels 17 to 24, and vibration signal channels 25 to 32. The sampling trigger times of each channel are sequentially delayed, ensuring that the trigger actions are evenly staggered within the same sampling period. The trigger time for channel 1 is set to 0 microseconds, for channel 2 to 3 microseconds, for channel 3 to 6 microseconds, and so on, with the trigger time for channel 32 set to 93 microseconds. The sampling window length for each channel remains 1 microsecond to ensure no overlap between sampling points. After all channels complete sampling within the same period, the entire sampling period forms a continuous staggered structure from 0 to 100 microseconds, separating the signals from different channels on the time axis. This staggered distribution method ensures that only one channel is in sampling state at any given time, thereby avoiding overlap between sampled signals at the physical time level and laying the foundation for high-frequency signal separation.

[0059] After the initial allocation of sampling time slots, the sampling rhythm needs to be determined based on the dominant frequency characteristics of each channel signal, ensuring that each signal exhibits a non-overlapping sampling rhythm on the time axis. Taking common substation signals as an example, the dominant frequency of voltage signals is 50Hz, current signals are 2000Hz, temperature signals have a variation frequency below 10Hz, and vibration signals have a dominant frequency of approximately 800Hz. Appropriate sampling interval differences are set for the characteristics of different signal types. The sampling interval for low-frequency signals is maintained at 3 microseconds, for medium-frequency signals at 6 microseconds, and for high-frequency signals at 9 microseconds. Within one sampling period, the voltage signal has 333 sampling points, the current signal has 166 sampling points, the temperature signal has 111 sampling points, and the vibration signal has 55 sampling points. Through this sampling interval design, the sampling density of high-frequency signals is sparser than that of low-frequency signals, and different types of signals form a fixed interleaved structure in the time domain. For example, within the same millisecond time range, voltage signals are sampled at 0, 3, and 6 microseconds; current signals at 1.5, 7.5, and 13.5 microseconds; vibration signals at 4.5, 10.5, and 16.5 microseconds; and temperature signals at 9, 18, and 27 microseconds. This ensures that the sampling trigger time for each signal type is within a different time window. Compared to existing uniform sampling rate methods, this rhythmic structure achieves physical separation of the signal layer in the time dimension, avoiding energy superposition caused by simultaneous acquisition of high-frequency channel signals at the same time.

[0060] After the sampling rhythm was rearranged, a staggered sampling spectrum was generated using continuous time scale bands to verify the independence of the signal in the time and frequency domains. The staggered sampling spectrum uses the time scale as the horizontal axis and the sampling trigger time of each channel as the vertical axis to reflect the signal distribution density in the time domain. During the plotting process, sampling points were distinguished by color according to channel number: red for voltage signals, blue for current signals, green for temperature signals, and orange for vibration signals. On the time axis, it can be clearly observed that the four types of signals are periodically staggered, with no overlapping areas of sampling points. To verify the frequency domain effect, the sampling data of each channel was converted into a frequency distribution. The peak values ​​of the voltage signal are concentrated in the 50Hz to 55Hz range, the peak values ​​of the current signal are distributed in the 1900Hz to 2100Hz range, the peak values ​​of the vibration signal are located in the 780Hz to 820Hz range, and the peak values ​​of the temperature signal are located in the 5Hz to 8Hz range. Analysis showed that the frequency peaks of each signal had no overlapping portions, indicating that time-domain staggered sampling achieved frequency-domain signal separation. Further calculations of the cross-correlation coefficients between the signals revealed that the cross-correlation coefficient between voltage and current signals was 0.01, between current and vibration signals was 0.02, and between temperature and voltage signals was 0.005. These values ​​are all close to zero, demonstrating that there is no coupling between different signal channels in the time sampling sequence. This staggered sampling spectrum serves as both a visualization of the time separation and a quantitative basis for judging the effectiveness of frequency separation.

[0061] To ensure the stability and repeatability of the staggered sampling spectrum during long-term operation, dynamic verification and drift compensation of the sampling rhythm are required. The verification cycle is set to be performed every 10 seconds, and each verification compares the deviation of the current channel sampling interval from the standard interval. When a sampling interval drift of more than 0.1 microseconds is detected in any channel, rhythm compensation is performed. The compensation process is achieved by adjusting the delay of the sampling trigger signal. When the sampling delay of channel 3 is 0.15 microseconds, the trigger delay is reduced by 0.15 microseconds to restore the predetermined sampling time. To verify the compensation effect, 10 consecutive sets of sampling rhythm tests were performed, each set containing 1000 sampling points, and the average deviation of the sampling interval was recorded. The results show that after compensation, the average deviation of all channels does not exceed 0.03 microseconds, and the maximum deviation does not exceed 0.07 microseconds. Further monitoring of spectrum stability showed that after one hour of continuous operation, the peak frequency fluctuation of the current signal was 0.08Hz, and the peak frequency fluctuation of the vibration signal was 0.05Hz, with no spectrum drift observed. The verification and compensation mechanism ensures that the time structure of the staggered sampling spectrum remains stable during long-term operation, making the signal separation effect effective in the long term and providing high-precision time input for subsequent mirror trajectory plotting.

[0062] Identifiable weak marker pulses are injected into key sampling nodes of the staggered sampling spectrum. The reflected path of the marker pulses in the frequency domain is used to draw the mirror trajectory and generate a mirror trajectory map to locate the aliasing region in the signal spectrum.

[0063] After constructing the staggered sampling spectrum, to accurately identify the aliasing regions formed by frequency reflection in the sampled signal, it is necessary to inject weak, amplitude-controlled marker pulses at key sampling nodes in the time domain. Based on the reflection paths of these pulses in the frequency domain, a reflection mirror trajectory is plotted, forming a mirror trajectory diagram. The mirror trajectory diagram is used to clearly define the specific location and range of the aliasing region in the spectrum, thus providing a precise basis for subsequent signal phase inversion suppression and restoration of the true signal channel. The specific steps are as follows:

[0064] Key sampling nodes were identified in the staggered sampling spectrum. The staggered sampling spectrum consists of the time distribution of multiple acquisition channels, each with an independent sampling trajectory on a continuous time scale. Using a 1-millisecond sampling period as a baseline, this period was divided into 1000 time scale points, each spaced 1 microsecond apart. To ensure that the marker pulse forms a stable reflection trajectory in the frequency domain, points with high signal energy density and stable time variation were selected as injection nodes. Taking a current signal as an example, its main frequency distribution is between 2000 and 5000 Hz, with the corresponding sampling times concentrated between 120 and 200 microseconds of each period. Within this time period, nodes were selected where the instantaneous amplitude of the signal is greater than 1.5 times the average amplitude of the channel and the phase change between adjacent sampling points does not exceed 10 degrees. After screening, 120 microseconds, 160 microseconds, and 200 microseconds were selected as key sampling nodes. The signals at these nodes are in a high-energy and stable-phase region, ensuring that the injected pulse can form a clear reflection peak in the frequency domain without being masked by noise or interference from the original signal.

[0065] A identifiable weak marker pulse is injected at the identified key sampling nodes. The energy of the marker pulse must be small enough to avoid disturbing the original signal, yet large enough to form an independent reflection feature in the frequency domain. Taking a voltage signal peak of 500 volts and a current signal peak of 50 amps as examples, the pulse amplitude is set to 5 volts and 0.5 amps respectively, with the pulse energy accounting for approximately 2% of the average energy of the original signal. The pulse width is set to 1 microsecond, with the rise slope and fall slope controlled at 2 volts per nanosecond. The pulse interval is 40 microseconds, corresponding to the node time interval determined in the previous step. Pulse injection uses a time superposition method, directly injecting electrical or current signals into the sampling signal channel at specified time scales, maintaining strict synchronization with the sampling rhythm. Taking the current channel as an example, the injection pulse times are 120 microseconds, 160 microseconds, and 200 microseconds, with a pulse amplitude of 0.5 amps for each time. After injection, a regular sequence of micro-amplitude spikes is formed in the time domain, which will generate corresponding reflection components in the frequency domain. By repeatedly injecting the signal for 10 consecutive sampling cycles, sufficient statistical strength of the pulse reflection signal can be ensured during frequency domain analysis. Unlike traditional methods that rely solely on static spectrum comparison to determine aliasing, this process actively injects low-energy signals to create markers, making the frequency domain features identifiable and traceable.

[0066] A frequency domain transformation was performed on the sampled signal containing the marked pulses to analyze the reflection path of the marked pulses in the spectrum and to plot the reflected mirror trajectory. The sampling window length was set to 10 milliseconds, containing continuous data for 10 complete sampling cycles. The data was mapped to a frequency range of 0 to 10000 Hz, and a spectrum was plotted with frequency on the x-axis and signal amplitude on the y-axis. In the spectrum, the marked pulses formed several discrete narrowband peaks, with the frequencies corresponding to the peaks corresponding one-to-one with the time nodes. Taking the current signal channel as an example, three clear peaks were observed in the 2000 to 5000 Hz frequency band, located at 2100 Hz, 2500 Hz, and 2900 Hz, respectively. The phase difference between each peak and the original signal peak was measured, revealing a phase difference of 180 degrees for the 2100 Hz peak, 135 degrees for the 2500 Hz peak, and 90 degrees for the 2900 Hz peak, indicating a reflection phenomenon. By linearly fitting the phase changes and energy reduction laws of these reflection peaks, the reflected mirror trajectory curves were obtained. The curve is relatively flat at the low-frequency end, but exhibits a distinct bend at the high-frequency end, with its radius of curvature gradually decreasing from 45 Hz / µs to 15 Hz / µs, indicating that the high-frequency components undergo more severe foldback. The location of the foldback point is directly related to the signal sampling interval and pulse injection time, and is an important basis for subsequent localization of aliasing regions.

[0067] A mirror trajectory diagram was drawn based on the reflected mirror trajectory, and the range of the aliasing region was confirmed by the spectral energy distribution. The mirror trajectory diagram uses time as the x-axis and frequency as the y-axis, mapping each reflection peak to a specific time point. Using 2100 Hz corresponding to 120 microseconds, 2500 Hz to 160 microseconds, and 2900 Hz to 200 microseconds as coordinate points, a continuous curve was formed by connecting these points. The energy density in the reflection point region of the curve was significantly higher than in other regions, indicating the presence of aliasing signals in this region. The rate of curvature change was calculated, and an aliasing region was defined when the curvature change was greater than 15%. Energy integration in this region showed that it accounted for 7% of the overall spectral energy, with high-frequency components accounting for 5% and low-frequency components accounting for 2%. Comparison with the spectrum of the uninjected pulse signal showed that the energy difference in the aliasing region was only 0.5%, indicating high aliasing localization accuracy. The aliasing region was marked in the time domain as 120 to 200 microseconds, corresponding to the frequency range of 2100 to 2900 Hz. To further verify accuracy, statistics were compiled for 50 consecutive sampling periods. The fluctuation in the aliasing region position was less than 2 microseconds, and the frequency fluctuation was less than 10 Hz, demonstrating that the method has good repeatability and stability. The mirror trajectory diagram not only visually displays the distribution of the aliasing region but also provides a dynamic adjustment basis for subsequent high-frequency signal suppression. When high-frequency interference or load switching occurs in the substation operating conditions, the reversal point position of the mirror trajectory will undergo a measurable shift. By monitoring the trajectory shift in real time, the sampling rhythm can be automatically corrected to maintain spectral stability.

[0068] Based on the mirror trajectory map, an anti-phase suppression window is set in the aliasing contamination frequency band. By weakening the energy of the mirror component at the edge of the window, the real high-frequency signal channel is restored, and a list of real signal channels is formed.

[0069] After obtaining the mirror trajectory map and identifying the aliasing contamination frequency band, in order to remove the interference of the reflected mirror components, restore the true high-frequency signal channel, and form a stable signal list, it is necessary to establish an anti-phase suppression window based on the energy distribution results of the mirror trajectory map. This window weakens the mirror energy through energy inverse cancellation and edge buffer control, thereby redistributing the energy gradient in the contaminated signal region. The specific implementation steps are as follows:

[0070] The frequency range and time boundary of the phase suppression window are determined. The mirror trajectory plot, drawn in the previous step, reflects the precise location and energy density distribution of the aliasing region. Taking a current signal as an example, the mirror trajectory curve exhibits a clear zigzag pattern between 2100 Hz and 2900 Hz. The energy density in this range is approximately 40% higher than adjacent frequency bands, indicating that this frequency band is the main aliasing region. To suppress this region, the center frequency of the phase suppression window is set at 2500 Hz, the frequency band range is 2050 Hz to 2950 Hz, and the total width is 900 Hz. The frequency band is divided into three equal-width sub-segments: the first sub-segment ranges from 2050 to 2350 Hz, the second sub-segment ranges from 2350 to 2650 Hz, and the third sub-segment ranges from 2650 to 2950 Hz. Each sub-segment corresponds to three main return peaks in the mirror trajectory curve, with peak positions at 2100 Hz, 2500 Hz, and 2900 Hz, respectively. To prevent signal discontinuity caused by abrupt energy changes at the window edges, a smooth transition zone of 50 Hz is added on both sides of the window, with an inner attenuation rate of 0.5% per Hz and an outer attenuation rate of 0.3% per Hz, thus forming a smooth energy transition band. The window boundary position maintains a real-time correspondence with the frequency center in the mirror trajectory diagram. When a trajectory curve offset exceeding 20 Hz is detected, the window center frequency is automatically adjusted. For example, if the aliasing center frequency drifts from 2500 Hz to 2520 Hz, the window center frequency is synchronously adjusted to 2520 Hz to prevent misalignment of the suppression region. This dynamic boundary adjustment strategy ensures that the window and the aliasing region overlap, guaranteeing stable subsequent suppression effects without loss of effective signal.

[0071] Inversion suppression is performed within a defined frequency band to attenuate the mirror energy peak and restore spectral balance. This process achieves mirror energy cancellation by superimposing a phase-reversed signal within the aliasing band. The phase of the inverted signal differs from the mirror signal by 180 degrees, its amplitude is 0.95 times the peak value of the mirror, and its frequency points strictly correspond to the mirror energy peak. Using three reflected peaks as references, inverted signals are injected at three frequency positions: 2100 Hz, 2500 Hz, and 2900 Hz. The duration of the inverted signal is 0.5 ms, with amplitude reduction ratios of 100%, 80%, and 60%, respectively, to form a reduction gradient from the center to the edge. To prevent the attenuation operation from accidentally damaging the real high-frequency signal, an energy protection band is set at the edge of the window. The band width is 30 Hz, with an inner reduction ratio of 50% and an outer reduction ratio of 20%, forming a buffer energy transition region. After the phase inversion operation, the energy peaks of the mirror spectrum decreased significantly. Measurements showed that the peak amplitude at 2100 Hz decreased from 12 mV to 0.8 mV, at 2500 Hz from 10 mV to 0.6 mV, and at 2900 Hz from 8 mV to 0.4 mV, with an average suppression rate exceeding 93%. After energy attenuation, the overall spectrum remained smooth and continuous without any new frequency abrupt changes. To verify the uniformity of the suppression, the energy within the 2050 to 2950 Hz band was integrated. The results showed that the energy before suppression was 7.2 mW, and after suppression it was 0.6 mW, a difference of 6.6 mW, indicating that the energy of the mirror component was completely attenuated while the high-frequency signal remained undamaged. Unlike traditional fixed band-stop filtering methods, this phase inversion suppression process achieves symmetrical energy cancellation through phase inversion control, maintaining a consistent spectral shape and preventing the true signal from being attenuated.

[0072] High-frequency signals after phase inversion suppression were extracted, and the true channels were restored to form a signal channel list. After the suppression operation, the true high-frequency signals that were masked by the mirrored energy reappeared in the spectrum. Energy scanning analysis revealed a significant increase in energy density in the range of 2900 Hz to 5000 Hz, with the peak value increasing from 1.2 mW to 2.1 mW, an increase of 75%. Based on this, each frequency channel was screened for both energy and phase, defining the frequency points with signal amplitudes higher than 1.2 times the average amplitude and phase fluctuations less than 0.05 radians as true channels. After screening, 45 effective channels were identified, of which 71% had energy intensities greater than the benchmark value. Each channel was arranged in ascending order of frequency, and the channel number, frequency range, amplitude value, phase offset, and time scale position were recorded. For example, channel 1 has a frequency range of 2950 to 3100 Hz, channel 2 ranges from 3100 to 3250 Hz, channel 3 ranges from 3250 to 3400 Hz, channel 4 ranges from 3400 to 3550 Hz, and channel 5 ranges from 3550 to 3700 Hz. Stability verification was performed on each channel for 5 sampling periods. The results showed that the frequency fluctuation was less than 15 Hz and the amplitude change did not exceed 2%. This resulted in a list of real signal channels, recording the time-frequency attributes of each channel, providing a precise data source for subsequent early warning anchor point extraction. This list visually displays the real channel distribution of high-frequency signals, enabling accurate identification of early anomalies in electrical equipment.

[0073] The arrangement order of the feature signal channels is reconstructed based on the real signal channel list, and the real high-frequency signal segments are concentrated into a unified identification sequence to generate a set of early warning anchor points, which are used as the core input in the dynamic monitoring stage.

[0074] After obtaining the actual list of signal channels, to ensure a unified identification structure and temporal continuity for multi-channel signals during the dynamic monitoring phase, it is necessary to reconstruct all characteristic signal channels in the list, rearranging the order of each channel so that high-frequency signal segments are continuously distributed along the time axis and form an ordered identification sequence in the frequency dimension. After reconstruction, a set of early warning anchor points is generated based on energy mutation points and phase change points to provide accurate input data for the dynamic monitoring phase. The specific implementation steps are as follows:

[0075] The reconstruction priority of channels was determined based on a list of real signal channels. This list recorded the frequency range, amplitude, phase offset, time scale position, and energy stability indicators for each channel. Taking the 45 high-frequency channels in the list as an example, their frequency range is from 2950 Hz to 5000 Hz. To ensure high energy concentration and phase stability during signal reconstruction, priority ranking was set based on two dimensions: average energy density and phase fluctuation amplitude. The average energy density of each channel was calculated by measuring energy over five consecutive sampling periods, in milliwatts; the phase fluctuation amplitude was obtained by calculating the phase difference between adjacent sampling points, in radians. Channels with an energy density greater than 1.3 times the average and a phase fluctuation less than 0.05 radians were selected as first-level channels; channels with an energy density between the average and 1.3 times the average and a phase fluctuation less than 0.08 radians were selected as second-level channels; and the remaining channels were third-level channels. Actual statistical results showed 17 first-level channels, 18 second-level channels, and 10 third-level channels. The frequency range of the first-level channels is concentrated between 2950 Hz and 3800 Hz, with an average energy density of 2.1 mW and a phase stability better than 0.03 radians. The second-level channels are mainly distributed in the 3800 Hz to 4500 Hz range, with an average energy density of 1.6 mW and a phase stability of 0.05 radians. The third-level channels are distributed in the 4500 Hz to 5000 Hz range, with an energy density of 1.1 mW and a phase stability of 0.07 radians. Based on the above classification results, the signal reconstruction priority is determined, with the first-level channels prioritized, followed by the second-level channels, and then the third-level channels. This priority determination mechanism ensures that high-energy, phase-stable signal segments are aggregated first during signal reconstruction, resulting in the final identified sequence with the highest signal-to-noise ratio and structural stability.

[0076] Feature channels are arranged and reconstructed according to priority, concentrating real high-frequency signal segments into a continuous recognition sequence. The reconstruction process uses a time scale as a reference, realigning the start and end points of each channel's signal on the time axis to ensure uninterrupted temporal splicing. A 1-millisecond reconstruction window is used, arranging the signal segments of each channel sequentially in chronological order. The first-level channel signal is placed at the beginning of the recognition sequence, covering a time range of 0 to 0.35 milliseconds; the second-level channel signal is placed in the middle, covering a time range of 0.35 to 0.75 milliseconds; and the third-level channel signal is placed at the end, covering a time range of 0.75 to 1.0 millisecond. To ensure continuity between adjacent channel signals at frequency transitions, a 50 Hz cross-fusion band is set at the channel junctions. Taking channel 1 (2950 to 3100 Hz) and channel 2 (3100 to 3250 Hz) as examples, a frequency overlap region is established at 3100 Hz, with an energy distribution ratio of 60% for the first channel and 40% for the second channel. Frequency transition is achieved smoothly through energy weighting. During channel splicing, phase consistency must also be ensured. If the phase difference between adjacent channels is greater than 5 degrees, a phase adjustment is applied to the signal of the second channel, with the adjustment amplitude being half of the difference, ensuring phase continuity at the transition point. After complete reconstruction, the identified sequence forms a seamless distribution on the time axis, the overall signal energy density is increased to 1.9 times the original mean, the peak amplitude is increased by approximately 70%, and the signal continuity index reaches 98%. To verify the reconstruction effect, spectral analysis is performed. The frequency distribution is continuous and uninterrupted from 2950 Hz to 5000 Hz, with an average peak spacing of 150 Hz and an energy difference of less than 5% between adjacent frequency bands, indicating that the signal after channel fusion is generally stable and energy-balanced.

[0077] A set of early warning anchor points is generated based on the reconstructed recognition sequence. The determination of early warning anchor points uses both signal energy abrupt changes and phase changes as dual criteria. By analyzing the temporal waveform of the recognition sequence, the rate of energy change between adjacent time periods is detected; when the rate of change is greater than 15%, it is marked as an energy anchor point. Simultaneously, the phase curve is detected; when the phase change exceeds 10 degrees, it is marked as a phase anchor point. The positions of the two types of anchor points are cross-compared; an energy anchor point and a phase anchor point with a time position difference of less than 3 microseconds are considered valid early warning anchor points. With a detection period of 10 milliseconds, a total of 65 energy anchor points and 47 phase anchor points were detected, forming 22 valid early warning anchor points. Each early warning anchor point includes five parameters: time position, frequency center, signal amplitude, energy gradient, and phase offset. For example, the first anchor point has a time position of 140 microseconds, a frequency center of 3050 Hz, a signal amplitude of 1.9 milliwatts, an energy gradient of 18%, and a phase shift of 8 degrees; the second anchor point has a time position of 180 microseconds, a frequency center of 3250 Hz, a signal amplitude of 2.0 milliwatts, an energy gradient of 21%, and a phase shift of 10 degrees; the third anchor point has a time position of 220 microseconds, a frequency center of 3400 Hz, a signal amplitude of 2.2 milliwatts, an energy gradient of 23%, and a phase shift of 9 degrees. To verify the stability of the anchor points, monitoring was conducted for 20 consecutive sampling periods. The anchor point time position fluctuation was less than 2 microseconds, the frequency shift was less than 5 Hz, and the energy gradient change was less than 3%, proving that the anchor point set has high repeatability. By arranging the anchor points in chronological order, a warning anchor point set is formed. This anchor point set reflects the dynamic changes of high-frequency signals in the time and frequency dimensions and is a key input for judging equipment overload, insulation aging, contact arcing, and abnormal vibration during the dynamic monitoring phase. Unlike existing alarm triggering methods based on fixed thresholds, this early warning anchor point set is generated through the dynamic evolution of the signal itself, possessing proactive prediction and adaptive adjustment capabilities, and can provide early risk warnings when early characteristic changes occur in the device.

[0078] A time-reversal phase traction operation is performed around the early warning anchor point set. A miniature phase curtain is laid in front of the early warning anchor point. Combined with an energy bypass structure and adjustable virtual impedance, phase inverse traction and energy migration are performed, thereby constructing a dynamic control closed loop with stable spectrum and constant early warning threshold.

[0079] After forming the early warning anchor point set, in order to maintain spectral stability and a constant early warning threshold for the high-frequency signal during the dynamic monitoring phase, a time-reversal phase traction operation needs to be performed around the early warning anchor point set. This is achieved by laying a miniature phase curtain in front of the anchor points, establishing an energy bypass structure, and introducing an adjustable virtual impedance, thereby realizing signal phase reversal and energy transfer, ultimately forming an adaptive closed-loop dynamic control process. The specific steps are as follows:

[0080] A time-reversal trigger operation is performed, generating a backpropagation signal using a set of warning anchor points as the core reference. The warning anchor point set contains multiple marker points with time-frequency attributes; each anchor point records its time position, frequency center, amplitude, phase offset, and energy gradient information. For example, anchor point 1 has a time position of 140 microseconds, a frequency center of 3050 Hz, a signal amplitude of 1.9 milliwatts, and a phase offset of 8 degrees. To form the time-reversal signal, a time-reversal trigger window is set 5 microseconds ahead of the anchor point, with a window width of 10 microseconds, to capture the continuous signal segment before the anchor point. A reverse signal is generated within this window, with the same frequency as the anchor point signal, an amplitude 0.9 times that of the anchor point signal, and a phase delay of 180 degrees. For example, the reverse signal has a frequency of 3050 Hz, an amplitude of 1.71 milliwatts, and a phase delay of 180 degrees. The reverse signal propagates from 145 microseconds to 135 microseconds in time, in the opposite direction to the original signal, forming a local time-reversal interval. When the inverted signal encounters the original signal, energy interference occurs, forming a phase transition region that provides the physical basis for subsequent phase pulling and energy conduction. Unlike traditional methods that only correct signal errors in the frequency domain, this process directly constructs a backward propagation path on the time axis, achieving physical cancellation of signal disturbances at the source.

[0081] A miniature phase curtain is laid inside the time-reversal window to precisely control the phase distribution and propagation speed of the reverse signal. The miniature phase curtain is composed of a thin layer of highly conductive polymer, 0.15 mm thick, covering a 10-microsecond range within the time-reversal region, with phase modulation units evenly spaced along the time axis. The phase modulation units are spaced 0.5 microseconds apart, with a total of 20 modulation points. Each modulation point introduces a fixed phase delay along the reverse signal propagation path. Using a reverse signal frequency of 3050 Hz as a reference, the phase delay angle of a single modulation point is 360 degrees divided by the number of discrete sampling points corresponding to the signal period, resulting in a phase delay of approximately 5.3 degrees per point. The first modulation unit at the beginning of the phase curtain has an initial phase of 180 degrees, which decreases by 5.3 degrees at each subsequent point, reaching 0 degrees at the 20th modulation unit, achieving a smooth phase decrease. This process ensures that the phase of the reverse signal gradually synchronizes with the forward signal after passing through the phase curtain, transitioning from a completely reversed state to an in-phase state. To prevent reflections during phase transition, impedance matching layers are installed at both ends of the miniature phase curtain, with an input impedance to output impedance ratio of 1.02:1 to ensure continuous signal propagation. After phase curtain adjustment, the phase error between the reverse signal and the original signal is less than 1 degree at time 135 microseconds, and the energy superposition cancels each other out, completing the initial phase alignment.

[0082] An energy bypass structure is constructed behind a miniature phase curtain, allowing the inverted energy to migrate and dissipate along a specific path. The energy bypass structure consists of three layers of conductive channels and a set of energy dissipation elements. Each channel is 2 microseconds wide, has a 5-ohm on-resistance, and a 0.8 millihenry inductance. After the reverse signal passes through the phase curtain, its energy is diverted to the three channels. The first channel handles the extraction of high-frequency energy, ranging from 2950 to 3500 Hz; the second channel extracts mid-frequency energy, ranging from 3500 to 4200 Hz; and the third channel extracts low-amplitude energy, ranging from 4200 to 5000 Hz. This layered extraction disperses energy across frequency ranges, preventing localized overheating caused by energy concentration. Impedance matching resistors are installed at the end of each channel, with a matching value of 0.95 times the input impedance, ensuring that energy is transmitted to the dissipation elements without reflection. The dissipation elements are high-resistivity carbon film resistor arrays with a power capacity of 5 watts and a resistance temperature rise controlled within 0.3 degrees Celsius per second. Testing showed that after bypassing the inverted energy, the original signal energy decreased by 93%, and the peak amplitude of the spectrum decreased to 7% of its original value, indicating that energy transfer and dissipation were complete. This energy bypassing process not only reduces peak fluctuations in the spectrum but also maintains signal phase continuity, ensuring that energy flow and spectrum stability are synchronized.

[0083] An adjustable virtual impedance is introduced into the energy bypass path to precisely control the rate of phase reversal and energy transfer efficiency. The virtual impedance, formed by a combination of inductive and capacitive elements, enables dynamic phase compensation. The capacitor value ranges from 50 to 500 microfarads, and the inductor value ranges from 0.1 to 1.0 millihenries. Specific parameters are determined based on the anchor signal frequency; at 3050 Hz, a capacitor of 200 microfarads and an inductor of 0.5 millihenries are selected, forming an equivalent impedance of 36.7 ohms. The virtual impedance, connected in series with the energy bypass, creates a controllable phase delay during reverse signal propagation. Initially, the phase difference between the inverted signal and the original signal is 180 degrees. After virtual impedance adjustment, the phase difference gradually decreases to 10 degrees, ultimately achieving phase synchronization. To avoid signal reflection caused by impedance abrupt changes, a buffer delay of 5 microseconds is set during adjustment to ensure a smooth change in impedance value over time. At this point, the inverted energy gradually migrates along the virtual impedance channel to the energy balance region where the anchor signal is located. The energy flow direction changes from reverse propagation to forward traction, and the energy distribution tends to be balanced. After virtual impedance control, the signal energy achieves stable exchange within 1 microsecond before and after the anchor point, with an energy difference of no more than 3%, a phase difference of no more than 2 degrees, and the main peak position of the spectrum remains unchanged at 3050 Hz.

[0084] Closed-loop verification was performed on the signal after phase pulling and energy transfer to evaluate spectral stability and the constancy of the warning threshold. The test period was 10 milliseconds, and the detection frequency range was 2950 to 5000 Hz. Signal energy and phase changes were monitored, and the range of spectral peak fluctuations and the stability of the warning threshold were calculated. After 50 consecutive testing cycles, the peak energy fluctuation decreased from ±6% to ±1.1%, the phase drift decreased from 8 degrees to 0.8 degrees, and the warning threshold stabilized at around 2.0 mW with a fluctuation of 0.05 mW. Under long-term operation for 120 seconds, the spectral stability remained above 98%, the energy reflectivity was below 1%, and the phase cumulative error was less than 1 degree. Throughout the closed-loop process, the signal response delay was controlled within 2 microseconds, and the energy transfer efficiency reached 95%, indicating that spectral stability and energy conservation were achieved synchronously in a closed loop. Through the above process, time reversal, phase pulling, energy transfer, and virtual impedance control formed a collaborative working structure, enabling the signal to have self-adjusting capabilities under dynamic operating conditions.

[0085] The following is a complete example from a power field, which clearly explains the working process, key data and comparative effects of the invention in real working conditions. Finally, the core improvement indicators are summarized in a table.

[0086] In Zone B of a 110 kV substation, the main transformer has a capacity of 63 MVA and six important outgoing lines on the 10 kV side, supplying power to three metal rolling mill production lines, one large air compressor group, one electric arc furnace power supply circuit, and one comprehensive plant load. During peak weekday hours, the electric arc furnace frequently starts and stops between 18:30 and 19:30, resulting in concentrated harmonic outbursts; the rolling mill frequency converter performs batch load switching at 19:05, generating high-frequency disturbances. Traditional intelligent early warning systems rely on a uniform sampling rate and frequency domain threshold for judgment, which often results in spectral aliasing and mirror interference during this period, leading to false alarms and missed alarms. Over the past three months, an average of two false alarms occurred per week between 19:00 and 20:00. One instance of a cascading trip caused by a false alarm leading to protection logic lag resulted in an 18-minute power outage and direct economic losses of approximately 120,000 yuan.

[0087] After the invention is launched on this site, it operates according to the following link, and all of it is based on actual measured data:

[0088] First, a unified time anchor point was established and the rhythm was rearranged. All acquisition channels were aligned using a 10MHz high-stability clock, with a synchronization error stabilized at 0.15 microseconds. A sampling period of 1 millisecond was used to divide the time into 1000 time slots, with 32 channels triggering alternately at 3-microsecond intervals, forming a staggered sampling spectrum. At 19:05, the second of load switching, the main peak of the high-frequency current component was located in the 2000-5000 Hz range. After staggering, the cross-peak in the spectrum disappeared, and the cross-correlation coefficient decreased from 0.18 to 0.01. The historical average cross-correlation coefficient for this site was 0.16, which decreased to 0.02 under the same conditions after the implementation of this invention.

[0089] Secondly, weak marker pulses were injected into key nodes, and mirror trajectories were plotted. Marker pulses with an amplitude of 0.5 amps and a width of 1 microsecond were injected into selected current channels at 120 microseconds, 160 microseconds, and 200 microseconds. Three sets of identifiable discrete peaks appeared in the frequency domain at 2100 Hz, 2500 Hz, and 2900 Hz. Observed over a 10-millisecond window, the curvature of the mirror trajectory changed rapidly between 2400 and 2800 Hz, with a curvature change rate reaching 17%. The aliasing contamination frequency band obtained from the mirror trajectory mapping was 2100 to 2900 Hz, corresponding to a time range of 120 to 200 microseconds, with an energy percentage of 7%, of which the high-frequency aliasing component accounted for 5%.

[0090] Third, targeted decontamination and restoration of real channels using a reverse-phase suppression window. A 900 Hz suppression window was set around the mirror trajectory between 2050 and 2950 Hz, with an energy gradient transition at the edge of 50 Hz. Reverse energy with a phase difference of 180 degrees was injected into the three mirror peaks at 2100 Hz, 2500 Hz, and 2900 Hz, with an amplitude ratio of 0.95. After suppression, the amplitude of the mirror peaks decreased from 12 mV, 10 mV, and 8 mV to 0.7 mV, 0.5 mV, and 0.4 mV, respectively, and the proportion of contamination energy decreased from 7% to 1%. Forty-five real high-frequency channels were restored, covering frequencies from 2950 to 5000 Hz, with a phase stability better than 0.05 radians. Compared with the same shift on the same day before implementation, the number of real channels increased by approximately 1.7 times.

[0091] Fourth, channel reconstruction and early warning anchor point set generation. Based on the actual channel list, channels were rearranged using energy density and phase stability as dual indicators, and the channel peak spacing was balanced to 150 Hz. The energy density of the unified identification sequence after reconstruction increased from 0.8 mW to 1.6 mW. Statistical analysis over five consecutive sampling periods revealed 72 energy anchor points and 45 phase anchor points within the identification sequence, which intersected to form 22 effective early warning anchor points. Starting at 19:04:42, early indications of electric arc furnace switching showed phase abrupt changes near frequencies of 3250 Hz and 3400 Hz, with energy gradients of 21% and 23%, respectively, providing an early warning 24 seconds earlier than the actual ignition time of the electric arc furnace at 19:05:06. Thirteen seconds before the mill load switch at 19:08:15, an early warning anchor point with an energy gradient of 18% appeared near 3050 Hz, which was subsequently verified on-site to show bearing temperature rise and current pulsation.

[0092] Fifth, time-reversal phase traction and dynamic threshold stabilization closed loop. A time-reversal window and a miniature phase curtain are set 5 microseconds before each warning anchor point. Taking an anchor point time of 140 microseconds and a frequency of 3050 Hz as an example, a reverse signal with an amplitude of 1.71 mW and a phase delay of 180 degrees is generated, and the phase is gradually pulled from 180 degrees to 0 degrees through 20 phase modulation points. Subsequently, a three-layer energy bypass channel directs the reversed energy to an inductive dissipation link. After bypassing, the main peak energy fluctuation is reduced from ±6% to ±1.2%. With an adjustable virtual impedance of 200 μF and a ratio of 0.5 mH, the phase difference is compressed from 10 degrees to 2 degrees, and the energy difference within 1 microsecond before and after the anchor point does not exceed 3%. Hourly observations from 19:00 to 20:00 show that the main peak position of the spectrum drift is less than 5 Hz, and the fluctuation of the warning threshold of 2.0 mW is controlled within 0.05 mW. Compared with before implementation, the standard deviation of the threshold drift is reduced by approximately 82%.

[0093] The results of a week of continuous operation at the site showed that: false alarms decreased from twice a week to 0-1 times, with most occurring at the moment of new load commissioning; there were 0 missed alarms; the advance detection time was significantly improved compared to the threshold method, with electric arc furnace-related risks being alerted an average of 19 to 26 seconds in advance; and rolling mill bearing abnormalities were alerted an average of 12 to 18 seconds in advance. More importantly, no interlocking trips occurred, and protection delays caused by misoperation and misjudgment were zero during this week. Simultaneously, infrared thermography showed that the peak oil temperature of the main transformer decreased by 1.8 degrees Celsius during peak hours, the total harmonic distortion rate of the 10 kV bus voltage decreased from 4.7% to 3.2% between 19:00 and 20:00, and the estimated total active power loss decreased by approximately 2.3%.

[0094] To eliminate randomness, data from 19:00 to 20:00 on three comparable historical days at the same site were selected for comparison: the traditional method achieved a 61% success rate in identifying high-frequency features under the superimposed conditions of electric arc furnace and rolling mill, while the method of this invention improved this to 92%; the traditional false alarm rate was 7.5%, while the method of this invention was 1.2%; the early warning lead time was improved from less than 5 seconds to an average of 21 seconds; the variance of the main peak of the spectrum was reduced from 38.6 to 6.1; and the number of manual review work orders was reduced by about 54%, saving about 5.4 hours of review time per week, based on 15 minutes of manual work per order.

[0095] The table below summarizes the comparison results of the core indicators (using the average or median of a week's statistics):

[0096] Indicator Name Traditional numerical methods Numerical values ​​of the method of the present invention Increase or decrease Time synchronization error in microseconds 2.7 0.15 Reduced by 94.4% Spectral aliasing energy ratio 7.0% 1.0% Reduced by 85.7% Actual number of high-frequency channels 26 45 An increase of 73.1% Effective number of early warning anchor points 8 22 An increase of 175.0% Average lead time for early warning (seconds) 4.8 21.0 An increase of 337.5% False alarm rate 7.5% 1.2% Reduced by 84.0% underreporting rate 5.8% 0.0% Reduced by 100.0% Main peak energy fluctuation range ±6.0% ±1.2% Reduced by 80.0% Warning threshold drift standard deviation milliwatts 0.28 0.05 Reduced by 82.1% Frequency main peak variance 38.6 6.1 Reduced by 84.2% Total harmonic distortion rate of 19-20 4.7% 3.2% Reduced by 31.9% Number of interlocking trips per week 0.5 0.0 Reduced by 100.0% Review work order quantity / week 24 11 Reduced by 54.2% Direct power outages result in losses of tens of thousands of yuan per week. 4.0 0.0 Reduced by 100.0%

[0097] As can be seen from the above examples, this invention transforms high-frequency interference into controllable energy migration and phase alignment through a continuous link of time anchoring, staggered sampling, marked pulses, mirrored trajectories, phase inversion suppression, channel reconstruction, and time inversion phase traction. This achieves spectrum stability and constant threshold, significantly reducing false alarms and missed alarms, and advancing early warning from passive response to proactive early perception. It is suitable for complex substation scenarios where electric arc furnaces, frequency converters, and impulsive loads coexist.

[0098] This invention constructs a full-link early warning mechanism that achieves source synchronization, process separation, feature calibration, and stable output by establishing a unified time anchor network, recompiling sampling rhythms, injecting identification markers, suppressing mirror aliasing, reconstructing identification sequences, and performing time-reversal phase traction operations. Compared with existing techniques that rely on fixed frequency domain features and single threshold judgments, this invention can accurately identify hidden high-frequency feature signals masked by aliasing in the context of multi-source high-frequency disturbances, detect abnormal signs in advance, and greatly reduce the probability of false alarms and missed alarms. In continuous dynamic monitoring, the system can maintain stable spectral peaks, clear identification paths, and constant early warning thresholds, effectively avoiding protection malfunctions, identification delays, and equipment damage caused by false spectral feature interference, and significantly improving the intelligent identification capability and proactive defense level of substation operating status.

[0099] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A data analysis-based intelligent early warning method for substations, characterized in that, Includes the following steps: Establish a unified time anchor network to perform microsecond-level time alignment of electrical signals from each acquisition channel during substation operation, forming a continuous time scale band to provide a time reference for subsequent signal rhythm arrangement; The sampling rhythm is rearranged based on the continuous time scale band, and staggered sampling intervals are set between each acquisition channel to separate high-frequency signals in the time domain and construct staggered sampling spectrum to avoid frequency overlap of high-frequency signals. Identifiable weak marker pulses are injected into key sampling nodes of the staggered sampling spectrum. The reflected path of the marker pulses in the frequency domain is used to draw the mirror trajectory and generate a mirror trajectory map to locate the aliasing region in the signal spectrum. Based on the mirror trajectory map, an anti-phase suppression window is set in the aliasing contamination frequency band. By weakening the energy of the mirror component at the edge of the window, the real high-frequency signal channel is restored, and a list of real signal channels is formed. The arrangement order of the feature signal channels is reconstructed based on the real signal channel list, and the real high-frequency signal segments are concentrated into a unified identification sequence to generate a set of early warning anchor points, which are used as the core input in the dynamic monitoring stage. By performing time-reversal phase traction operations around the early warning anchor point set, a miniature phase curtain is laid in front of the early warning anchor point. Combined with an energy bypass structure and adjustable virtual impedance, phase inverse traction and energy migration are performed, thereby constructing a dynamic control closed loop with stable spectrum and constant early warning threshold.

2. The intelligent early warning method for substations based on data analysis according to claim 1, characterized in that, The process of forming continuous time scale bands is as follows: A unified time reference is established. The reference clock signal is output by the high-stability temperature-compensated crystal oscillator in the main control device and transmitted to the time receiving unit of each acquisition channel via a bidirectional optical fiber link. The phase adjustment circuit is used to make the local clock signal and the reference clock signal phase consistent. Perform periodic time deviation verification by calculating the trigger time deviation by sampling voltage, current, temperature, operating current and power factor signals, and injecting a synchronization pulse packet to correct the sampling timing when the deviation exceeds the limit; Establish a characteristic time stamp sequence, embed equally spaced calibration pulses into the time axis of each acquisition channel, and verify the consistency of the time scale by matching the stamp intervals; The continuous verification process is performed, the time drift rate is recorded, and the synchronization accuracy is maintained through timing pulse phase compensation, thereby forming a stable continuous time scale band.

3. The intelligent early warning method for substations based on data analysis according to claim 2, characterized in that, The process of constructing the staggered sampling spectrum is as follows: Based on the continuous time scale, the sampling time of each acquisition channel is redistributed so that the sampling actions are staggered on the time axis and the sampling windows of each channel do not overlap. The sampling rhythm is determined by combining the main frequency characteristics of the signals from each acquisition channel, and the sampling interval difference is set according to the frequency range of voltage signals, current signals, temperature signals and vibration signals, so that different signal types form a fixed interlaced structure on the time axis; A staggered sampling spectrum is generated using a continuous time scale band, and the signal independence between the time domain and the frequency domain is verified by cross-correlation coefficient calculation. Periodically perform sampling rhythm verification and drift compensation, and adjust the sampling trigger delay to maintain the stability of the staggered sampling spectrum and the continuity of the time structure.

4. The intelligent early warning method for substations based on data analysis according to claim 3, characterized in that, The steps for generating a mirror trajectory diagram are as follows: In the staggered sampling spectrum, key sampling nodes are determined based on signal energy density and phase stability. Injection nodes are selected at positions where the instantaneous amplitude of the signal is more than twice the average amplitude and the phase change between adjacent sampling points does not exceed a fixed angle. At the identified key sampling nodes, weak marker pulses with controlled amplitude are injected. The pulse width remains constant, the rising and falling edges have the same slope, and the pulse injection is synchronized with the sampling rhythm to form a recognizable pulse sequence. A frequency domain transformation is performed on the sampled signal after the injection of the marker pulse, and the reflected mirror trajectory is plotted based on the phase change and energy reduction law of the pulse reflection peak. A mirror trajectory map is formed based on the reflected trajectory, and the specific location and range of the aliasing region are determined by the spectral energy distribution.

5. The intelligent early warning method for substations based on data analysis according to claim 4, characterized in that, The process of generating the actual signal channel list is as follows: Based on the energy distribution of the mirror trajectory diagram, the frequency range and time boundary position of the anti-phase suppression window are determined, the frequency band is divided into multiple equal-width sub-segments, and energy smooth transition zones are set on both sides of the window to prevent signal breakage. In a defined frequency band, phase inversion suppression is performed by superimposing a phase inversion signal within the aliasing band to weaken the mirror energy peak and setting an energy protection band at the edge of the window to maintain spectral balance. The high-frequency signal after phase inversion suppression is extracted, and the real channel is restored based on energy and phase selection to form a list of real signal channels with recorded channel frequency range, amplitude value, phase offset and time scale position.

6. The intelligent early warning method for substations based on data analysis according to claim 5, characterized in that, The process of generating the early warning anchor point set is as follows: The reconstruction priority of the channels is determined based on the actual signal channel list, and the channels are sorted according to the average energy density and phase fluctuation amplitude, and the signal reconstruction order is determined. The characteristic signal channels are arranged and reconstructed according to priority order. The signals of each channel are realigned on the time axis based on the time scale band, and a frequency cross-fusion band is set at the channel junction to maintain signal continuity. A set of early warning anchor points is generated based on the reconstructed recognition sequence. Energy anchor points and phase anchor points are determined by detecting the rate of change of signal energy and the amplitude of phase change. An effective set of early warning anchor points is generated based on the time position difference between the two types of anchor points.

7. The intelligent early warning method for substations based on data analysis according to claim 6, characterized in that, In the process of generating the early warning anchor point set, the determination of energy anchor points and phase anchor points is based on the energy change rate and phase change amplitude of adjacent time periods. When the time position difference between the two types of anchor points is less than the preset time threshold, it is determined to be a valid early warning anchor point. The time position, frequency center, signal amplitude, energy gradient and phase offset of each early warning anchor point are recorded as parameters to form a complete anchor point set.

8. The intelligent early warning method for substations based on data analysis according to claim 6, characterized in that, A time-reversal phase-pull operation is performed around the early warning anchor point set. A miniature phase curtain is laid in front of it, and phase inverse pull and energy transfer are performed in combination with an energy bypass structure and adjustable virtual impedance. The dynamic control closed loop steps are as follows: Execute the time reversal trigger operation, generate a backpropagation signal with the early warning anchor point set as a reference, and set a time reversal trigger window in front of the anchor point to form a local time reversal interval; A miniature phase curtain is laid inside the time reversal window. The phase distribution and propagation speed of the reverse signal are controlled by a phase modulation unit, and impedance matching layers are set at both ends of the phase curtain to keep the signal propagating continuously. An energy bypass structure is constructed behind the micro phase curtain, so that the inverted energy can migrate in layers along the conduction channel and dissipate through the energy dissipation element; An adjustable virtual impedance is introduced into the energy bypass path, and a dynamic phase compensation structure is formed by combining inductive and capacitive elements. Closed-loop verification is performed on the signal after phase pulling and energy transfer to evaluate spectral stability and the constancy of the warning threshold.

Images