A time-frequency hybrid anti-interference identification method and system for an electric energy meter

By combining fundamental wave active stripping and time-frequency morphological evolution characterization, the problem of identifying and processing weak non-stationary interferences in electricity meters is solved, thereby achieving accuracy and security in electricity metering and adapting to real-time changes in grid frequency and phase.

CN123115831APending Publication Date: 2026-11-17QINGDAO YINGLIDA NEW ENERGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611312349.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively distinguish and process weak non-stationary interference in electricity meters, leading to metering errors and misjudgments, and lack real-time adaptability to grid frequency and phase.

Method used

By combining fundamental wave active stripping with time-frequency morphological evolution, the electricity meter signal is collected, the fundamental wave component reconstructed signal is generated and differential operation is performed. Combined with time-frequency disturbance spectrum and disturbance feature extraction, a composite disturbance fingerprint is generated for interference type identification, and metering correction and active alarm are performed.

Benefits of technology

It improves the anti-interference capability of power metering, enhances the ability to identify and process non-stationary interference, ensures the accuracy and safety of metering, and can adapt to real-time changes in power grid frequency and phase.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115831A_ABST
    Figure CN123115831A_ABST
Patent Text Reader

Abstract

This invention discloses a time-frequency hybrid anti-interference identification method and system for electricity meters, belonging to the field of electricity metering technology. The method synchronously acquires voltage and current signals at the metering point, tracks the fundamental frequency, amplitude, and phase in real time, and digitally synthesizes a fundamental component reconstructed signal. The original electrical signal and the fundamental component reconstructed signal are differentially analyzed point-by-point to obtain a differential residual signal. The differential residual signal is then decomposed into a time-frequency disturbance spectrum, and disturbance morphological features and disturbance evolution features are extracted and jointly encoded into a composite disturbance fingerprint. An interference identification model is used to output the interference type identification result, and metering correction or proactive alarm is performed accordingly. This invention, through active fundamental frequency stripping and joint characterization of time-frequency morphological evolution, highlights, distinguishes, and handles weak non-stationary interference, improving the anti-interference capability and security of electricity metering.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of electricity metering technology, and in particular to a time-frequency hybrid anti-interference identification method and system for electricity meters. Background Technology

[0002] Smart meters are key devices at the end of the power grid for collecting electricity consumption data, settling transactions, monitoring loads, and analyzing abnormal electricity usage. Their voltage and current sampling links are constantly connected to complex power grid environments. With the widespread integration of frequency converters, rectifiers, charging infrastructure, distributed power sources, power line carrier communication equipment, and wireless communication devices, the metering point signal, in addition to the fundamental frequency and stable harmonics, may also be superimposed with short-duration pulses, narrowband frequency sweeps, periodic communication bursts, and maliciously injected signals. These types of interference are typically weak in energy, short in duration, and their frequency varies over time, easily causing instantaneous sampling deviations, metering algorithm mismatches, and errors in judging abnormal electricity usage.

[0003] Existing technologies typically employ time-domain filtering methods such as low-pass, band-pass, or notch filtering to suppress interference in known frequency bands, or use Fourier transform, harmonic analysis, and fixed-frequency energy statistics to identify abnormal frequency components. Some electricity meters also set alarm thresholds based on total harmonic distortion, peak factor, and zero-crossing offset. These methods struggle to balance useful signal fidelity and interference suppression when the interference band overlaps with the fundamental or harmonic bands, and fixed-frequency models are ill-suited to adapt to real-time fluctuations in grid frequency, amplitude, and phase. Global frequency domain analysis mixes frequency information across the entire analysis window, making it easy for the energy of short-term weak disturbances to be overwhelmed by strong fundamental waves and stable harmonics.

[0004] Current technologies for handling weak non-stationary interference still face challenges such as difficulty in removing the fundamental background, highlighting transient energy, and jointly characterizing the interference pattern and its temporal evolution. Relying solely on single-point amplitude, fixed-frequency energy, or total harmonic distortion rate makes it difficult to distinguish between normal harmonics, PLC communication bursts, and malicious pulse injections, easily leading to missed detections and misjudgments. Furthermore, there is a lack of closed-loop connection between the identification results and measurement corrections, proactive alarms, and model updates. Therefore, there is an urgent need for an anti-interference identification scheme that can proactively eliminate strong fundamental background masking and jointly utilize time-frequency patterns and evolutionary laws. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a time-frequency hybrid anti-interference identification method and system for electricity meters. By actively stripping the fundamental wave and jointly characterizing the time-frequency morphological evolution, it enables the highlighting, differentiation, and handling of weak non-stationary interference, thereby improving the anti-interference capability and security of electricity metering.

[0006] The above objectives can be achieved through the following approach: A time-frequency hybrid anti-interference identification method for electricity meters includes: acquiring voltage and current signals at the metering point of the electricity meter to generate an original electrical signal; performing real-time tracking of the fundamental component of the original electrical signal to obtain a set of real-time fundamental component parameters characterizing the dynamic changes of the fundamental wave; performing digital synthesis based on the real-time fundamental component parameters to generate a fundamental component reconstruction signal that matches the fundamental component in the original electrical signal; performing differential operation on the original electrical signal and the fundamental component reconstruction signal to obtain a differential residual signal highlighting non-fundamental components; performing time-frequency decomposition on the differential residual signal to generate a two-dimensional time-frequency disturbance spectrum; extracting disturbance morphology features characterizing the spatial distribution of interference energy and disturbance evolution features characterizing the temporal changes of interference energy based on the time-frequency disturbance spectrum; coupling the disturbance morphology features and the disturbance evolution features to generate a composite disturbance fingerprint; inputting the composite disturbance fingerprint into an interference identification model, which then performs classification processing to obtain an interference type identification result.

[0007] Optionally, the step of real-time tracking of the fundamental component of the original electrical signal to obtain a set of real-time fundamental component parameters characterizing the dynamic changes of the fundamental wave includes: performing orthogonal component generation and synchronous rotating coordinate system transformation on the original electrical signal to obtain a DC component containing fundamental wave information and an AC component containing harmonic and interference information; performing low-pass filtering on the DC component to obtain a smooth DC component that eliminates high-frequency fluctuations; and calculating the real-time frequency, amplitude, and phase of the original electrical signal based on the smooth DC component to generate the real-time fundamental component parameters.

[0008] Optionally, the step of extracting perturbation morphology features representing the spatial distribution of perturbation energy based on the time-frequency perturbation map includes: performing adaptive threshold segmentation and connected component analysis on the time-frequency perturbation map to identify and segment independent perturbation energy clusters; calculating the geometric moments, time-domain span, frequency-domain span, and principal direction of each perturbation energy cluster to obtain morphology descriptors representing shape, size, and direction; and combining the morphology descriptors to generate the perturbation morphology features.

[0009] Optionally, the step of extracting perturbation evolution features characterizing the temporal variation of perturbation energy based on the time-frequency perturbation map includes: performing cross-window correlation on the perturbation energy clusters within a continuous time window to construct the motion trajectory of the perturbation energy clusters; extracting dynamic parameters of the motion trajectory, including the cluster occurrence period, duration, frequency drift rate, and energy fluctuation coefficient, to obtain evolution descriptors; and combining the evolution descriptors to generate the perturbation evolution features.

[0010] Optionally, coupling the perturbation morphological features with the perturbation evolution features to generate a composite perturbation fingerprint includes: vector quantizing the morphological descriptors in the perturbation morphological features to generate a morphological codebook and a corresponding morphological code sequence; modeling the transition relationship between adjacent morphological codes based on the perturbation evolution features and the morphological code sequence to generate a state transition matrix describing the transition rules between morphologies; and jointly encoding the occurrence frequency of the morphological codebook, the state transition matrix, and the perturbation evolution features to generate the composite perturbation fingerprint.

[0011] Optionally, before inputting the composite perturbation fingerprint into the interference recognition model, the method further includes generating the interference recognition model by: acquiring an interference sample library containing various known interference signal samples; sequentially performing fundamental component real-time tracking, fundamental component reconstruction, differential residual extraction, time-frequency decomposition, perturbation feature extraction, and composite perturbation fingerprint generation on each interference signal sample in the interference sample library to obtain a sample fingerprint set; and using the sample fingerprint set and its corresponding known interference type labels to perform supervised learning training on a classifier to generate the interference recognition model.

[0012] Optionally, after obtaining the interference type identification result, the method further includes: matching and obtaining a target anti-interference strategy from a strategy library storing the mapping relationship between interference types and response strategies based on the interference type identification result; if the target anti-interference strategy is a metering correction, then reconstructing the interference waveform based on the disturbance energy clusters corresponding to the composite disturbance fingerprint in the time-frequency disturbance spectrum, and subtracting the interference waveform from the original electrical signal to obtain purified metering data; if the target anti-interference strategy is an active alarm, then generating alarm information including interference type, occurrence time, frequency range and classification confidence and reporting it to the main station system.

[0013] Optionally, the method further includes online updating of the interference identification model: classifying the composite perturbation fingerprint using the interference identification model and obtaining classification confidence; identifying the composite perturbation fingerprint with a classification confidence below a confidence threshold as an unknown perturbation fingerprint; performing online clustering analysis on the unknown perturbation fingerprint; when the number of samples in any cluster reaches a scale threshold, marking the cluster as a new interference label, and using the new interference label to update the interference sample library and retrain the interference identification model.

[0014] Optionally, the step of collecting voltage and current signals from the electricity meter metering point to generate the original electrical signal includes: acquiring a synchronous sampling clock generated by a high-precision clock source; using the synchronous sampling clock to synchronously sample the voltage channel and the current channel to obtain a synchronous voltage and current sequence containing the precise phase relationship between voltage and current; and combining the synchronous voltage and current sequence into a multi-dimensional time sequence according to a unified timestamp as the original electrical signal.

[0015] Based on the same inventive concept, this invention also provides a time-frequency hybrid anti-interference identification system for electricity meters. The system includes: a synchronous acquisition module for acquiring voltage and current signals at the metering point of the electricity meter to generate an original electrical signal; a fundamental wave tracking module for real-time tracking of the fundamental wave component of the original electrical signal to obtain a set of real-time fundamental wave component parameters characterizing the dynamic changes of the fundamental wave; a fundamental wave reconstruction module for digitally synthesizing based on the real-time fundamental wave component parameters to generate a fundamental wave component reconstruction signal that matches the fundamental wave component in the original electrical signal; and a residual extraction module for differentially analyzing the original electrical signal and the fundamental wave component reconstruction signal. The system performs calculations to obtain a differential residual signal that highlights the non-fundamental components; a time-frequency spectrum generation module is used to perform time-frequency decomposition on the differential residual signal to generate a two-dimensional time-frequency disturbance spectrum; a disturbance feature extraction module is used to extract disturbance morphological features and disturbance evolution features based on the time-frequency disturbance spectrum; a composite fingerprint generation module is used to couple the disturbance morphological features and disturbance evolution features to generate a composite disturbance fingerprint; and an interference identification module is used to input the composite disturbance fingerprint into an interference identification model for classification processing to obtain interference type identification results, and to perform measurement correction, active alarm, or online model update based on the interference type identification results.

[0016] Compared with the prior art, the present invention has the following advantages: This invention reconstructs the signal by real-time tracking of the fundamental frequency, amplitude, and phase of the power grid and digitally synthesizing the fundamental component. This allows the reconstructed signal to dynamically change with the actual state of the power grid, avoiding pseudo-residuals generated by fixed 50Hz or fixed amplitude-phase models when there is frequency drift and load changes.

[0017] This invention actively strips away the dominant fundamental background by performing point-by-point differential analysis between the original electrical signal and the reconstructed signal from the fundamental component. This makes the previously masked weak harmonics, transient pulses, and frequency drift interference the main variable components in the differential residual signal, thereby improving the detectability of non-stationary interference.

[0018] This invention constructs a composite perturbation fingerprint by combining the geometric shape of connected domains in the time-frequency perturbation spectrum with the evolution trajectory across time windows. This enables interference identification to simultaneously utilize point-like, line-like, and sheet-like spatial structures, as well as periodicity, duration, and frequency drift patterns, thereby enhancing the ability to distinguish between benign harmonics, PLC communication, and maliciously injected signals.

[0019] This invention connects the interference type identification results with metering correction, active alarm, and online updates for unknown interference, enabling the identification results to directly drive waveform purification and master station alarms. Furthermore, it updates the identification model after accumulating new interference samples, thereby improving the safety and long-term adaptability of the power metering system. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of a time-frequency hybrid anti-interference identification method for electricity meters according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of synchronous acquisition, real-time fundamental wave tracking, fundamental wave reconstruction and residual extraction in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram illustrating the generation of time-frequency perturbation spectra, perturbation morphology features, and perturbation evolution features according to an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram illustrating the composite perturbation fingerprint, interference identification, anti-interference strategy, and online update in an embodiment of the present invention.

[0024] Figure 5 This is a schematic diagram of the structure of a time-frequency hybrid anti-interference identification system for an electricity meter according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0026] Reference Figure 1 One embodiment of the present invention proposes a time-frequency hybrid anti-interference identification method for electricity meters. By actively stripping the fundamental wave and jointly characterizing the time-frequency morphological evolution, it can highlight, distinguish and handle weak non-stationary interference, thereby improving the anti-interference capability and security of electricity metering.

[0027] The method described in this embodiment specifically includes: S1. Collect the voltage and current signals at the electricity meter's metering point and generate the raw electrical signal; In one embodiment of the present invention, step S1 includes the following steps: Obtain the synchronous sampling clock generated by the high-precision clock source; The voltage and current channels are sampled synchronously using the synchronous sampling clock to obtain a synchronous voltage and current sequence containing precise phase relationships; The synchronous voltage and current sequences are combined into a multi-dimensional time sequence according to a unified timestamp to generate the original electrical signal.

[0028] Specifically, such as Figure 2 As shown, the voltage signal is acquired from the voltage sampling channel of the energy meter, and the current signal is acquired from the current sampling channel of the energy meter. The voltage and current sampling channels share the same synchronous sampling clock. The high-precision clock source uses a temperature-compensated crystal oscillator, a oven-controlled crystal oscillator, or a local clock source synchronized with the master station's time signal. The synchronous sampling clock is used to ensure that the voltage and current channels complete analog-to-digital conversion at the same sampling time, avoiding additional phase deviations caused by channel sampling delays. The analog-to-digital conversion bit width of the voltage and current channels is no less than 16 bits, and the cutoff frequency of the analog anti-aliasing filter is less than half of the sampling frequency.

[0029] Sampling period writing: ; In the formula, This indicates the sampling period between two adjacent synchronous sampling points, in seconds. This indicates the sampling frequency used by both the voltage and current channels, measured in Hertz.

[0030] No. Writing at each synchronous sampling moment: ; In the formula, Indicates the first Each synchronous sampling moment; Indicates the starting sampling time of the current analysis window; Indicates the sequence number of the synchronous sampling point, and ; This indicates the total number of synchronized sampling points within the current analysis window; Indicates the sampling period.

[0031] No. Writing a synchronous sampling record: ; In the formula, Indicates the first One synchronous sampling record; Indicates the voltage channel is in The voltage value collected at all times, in volts; Indicates the current path is in The current value collected at all times is in amperes.

[0032] Original electrical signal writing: ; In the formula, This represents the original electrical signal corresponding to the current analysis window; Indicates the first One synchronous sampling record; Indicates the sequence number of the synchronous sampling point; This indicates the total number of synchronous sampling points within the current analysis window.

[0033] The sampled voltage and current values ​​are amplitude corrected according to the gain coefficients of the voltage sampling channel, the current sampling channel, and the zero-point offset calibration parameters, respectively. The corrected voltage and current values ​​retain the same timestamp and are sequentially written into the original electrical signal according to the sampling point number, so that the subsequent fundamental wave tracking process can utilize the true phase relationship between voltage and current.

[0034] For example, taking a single-phase smart energy meter with a rated voltage of 220V and a rated current of 10A as an application scenario, the sampling frequency is taken as... The sampling period is: The starting sampling time of the analysis window is taken as... ,Will Substituting into the synchronous sampling time formula, we get: ;exist At that time, the voltage channel acquired Current channel acquisition The 1600th synchronous sampling record is as follows: The current analysis window length is 1 second, and the total number of sampling points is: Therefore, the original electrical signal is: The voltage and current channels share the same sampling time, and the voltage and current phase relationship will not be affected by the asynchronous sampling of the channels, providing accurate input for subsequent real-time tracking of the fundamental frequency, amplitude and phase.

[0035] S2. Perform real-time tracking of the fundamental component of the original electrical signal to obtain a set of real-time fundamental component parameters characterizing the dynamic changes of the fundamental wave. In one embodiment of the present invention, step S2 includes the following steps: The original electrical signal is subjected to orthogonal component generation and synchronous rotating coordinate system transformation to obtain a DC component containing fundamental wave information and an AC component containing harmonic and interference information. The DC component is subjected to low-pass filtering to obtain a smooth DC component with high-frequency fluctuations removed. The real-time frequency, amplitude, and phase of the original electrical signal are calculated based on the smoothed DC component, and the real-time fundamental component parameters are generated.

[0036] Specifically, such as Figure 2 As shown, the voltage and current sampling sequences from the original electrical signal are input into an orthogonal signal generator. The orthogonal signal generator uses a digital orthogonal filter, a Hilbert transform, or a second-order generalized integrator to generate the in-phase and quadrature voltage components corresponding to the voltage sampling sequences, and the in-phase and quadrature current components corresponding to the current sampling sequences, respectively. The in-phase voltage component is denoted as... The orthogonal components of the voltage are denoted as The in-phase component of the current is denoted as The orthogonal components of the current are denoted as A synchronous rotating coordinate system tracks the phase with the fundamental wave. As a rotation angle, the fundamental component is converted into a DC component with a lower rate of change, and harmonics and non-stationary interferences are converted into AC components that fluctuate with time.

[0037] Voltage direct-axis component written as: ; In the formula, Indicates the first The voltage direct-axis component of each sampling point, in volts; Indicates the first The voltage in-phase component at each sampling point, in volts; Indicates the first The voltage quadrature components at each sampling point are expressed in volts. Indicates the first The fundamental wave tracking phase at each sampling point is expressed in radians. Represents the cosine function; This represents the sine function.

[0038] Voltage quadrature axis component writing: ; In the formula, Indicates the first The voltage quadrature-axis component of each sampling point, in volts; Indicates the first The voltage in-phase component at each sampling point; Indicates the first The voltage quadrature components at each sampling point; Indicates the first Fundamental wave phase tracking at each sampling point; Represents the cosine function; This represents the sine function.

[0039] The direct-axis component of current is written as: ; In the formula, Indicates the first The current direct-axis component at each sampling point is expressed in amperes. Indicates the first The in-phase current component at each sampling point is expressed in amperes. Indicates the first The current orthogonal components at each sampling point are expressed in amperes. Indicates the first Fundamental wave phase tracking at each sampling point; Represents the cosine function; This represents the sine function.

[0040] Current cross-axis component writing: ; In the formula, Indicates the first The cross-axis component of the current at each sampling point is expressed in amperes. Indicates the first The in-phase current components at each sampling point; Indicates the first The current orthogonal components at each sampling point; Indicates the first Fundamental wave phase tracking at each sampling point; Represents the cosine function; This represents the sine function.

[0041] First-order low-pass filtering is applied to the voltage direct-axis component, voltage quadrature-axis component, current direct-axis component, and current quadrature-axis component. The low-pass filter smoothing coefficient is tuned according to the allowable rate of change of the grid fundamental wave, the duration of the interference, and the fundamental wave tracking response time, so that the amplitude and phase changes of the fundamental wave with a longer duration are preserved, while the pulses and high-frequency fluctuations with a shorter duration are suppressed.

[0042] Voltage smoothing direct-axis component writing: ; In the formula, Indicates the first The voltage smoothed direct-axis component of each sampling point is expressed in volts. Indicates the first The voltage direct-axis component of each sampling point; Indicates the first Smooth the direct-axis component of the voltage at each sampling point; This represents the low-pass filter smoothing coefficient, and .

[0043] Voltage smoothing cross-axis component writing: ; In the formula, Indicates the first The voltage smoothed quadrature-axis component of each sampling point is expressed in volts. Indicates the first The cross-axis component of the voltage at each sampling point; Indicates the first Smooth cross-axis components of voltage at each sampling point; This represents the low-pass filter smoothing coefficient.

[0044] Current smoothing direct-axis component writing: ; In the formula, Indicates the first The current smoothed direct-axis component at each sampling point is expressed in amperes. Indicates the first The direct-axis component of the current at each sampling point; Indicates the first Smooth the direct-axis component of the current at each sampling point; This represents the low-pass filter smoothing coefficient.

[0045] Current smoothing cross-axis components are written as: ; In the formula, Indicates the first The current smoothed quadrature-axis component at each sampling point is expressed in amperes. Indicates the first The cross-axis component of the current at each sampling point; Indicates the first The current smoothing cross-axis component at each sampling point; This represents the low-pass filter smoothing coefficient.

[0046] The normalized phase error is constructed by using the ratio between the voltage smoothed cross-axis component and the peak amplitude of the voltage fundamental wave.

[0047] Normalized phase error writing: ; In the formula, Indicates the first Normalized phase error at each sampling point; Indicates the first Smooth the direct-axis component of the voltage at each sampling point; Indicates the first The voltage smoothed quadrature axis component at each sampling point; the square root term represents the first... The peak amplitude of the fundamental voltage at each sampling point.

[0048] Phase error integral state writing: ; In the formula, Indicates the first The phase error integral state of each sampling point, in seconds; Indicates the first Phase error integral state at each sampling point; This represents the normalized phase error; This indicates the synchronous sampling period, in seconds.

[0049] Writing real-time frequency correction: ; In the formula, Indicates the first The real-time frequency correction of each sampling point relative to the rated fundamental frequency, in Hertz; This represents the phase error scaling factor, measured in Hertz. This represents the normalized phase error; This represents the integral coefficient of the phase error, in Hertz per second. This indicates the phase error integral state.

[0050] Real-time fundamental frequency writing: ; In the formula, Indicates the first The real-time fundamental frequency of each sampling point is expressed in Hertz. This indicates the rated fundamental frequency of the power grid, expressed in Hertz (Hz). This indicates the real-time frequency correction amount.

[0051] Fundamental wave tracking phase writing: ; In the formula, Indicates the first The fundamental wave tracking phase at each sampling point is expressed in radians. Indicates the first Fundamental wave phase tracking at each sampling point; Represents pi; Indicates the real-time fundamental frequency; This indicates the synchronous sampling period.

[0052] The peak amplitude of the fundamental voltage wave is written as: ; In the formula, Indicates the first The fundamental peak amplitude of the voltage at each sampling point, in volts; Represents the smoothed direct-axis component of the voltage; This represents the smoothed cross-axis component of the voltage.

[0053] The peak amplitude of the fundamental current wave is written as: ; In the formula, Indicates the first The peak amplitude of the fundamental current at each sampling point, in amperes; Represents the smoothed direct-axis component of the current; This represents the smoothed cross-axis component of the current.

[0054] The voltage fundamental phase offset is written as: ; In the formula, Indicates the first The voltage fundamental phase offset at each sampling point, in radians; Represents the arctangent function in the four quadrants; Represents the smoothed cross-axis component of voltage; This represents the smoothed direct-axis component of the voltage.

[0055] The fundamental phase offset of the current is written as: ; In the formula, Indicates the first The phase offset of the fundamental current at each sampling point, in radians; Represents the arctangent function in the four quadrants; Indicates the smoothed quadrature axis component of the current; This represents the smoothed direct-axis component of the current.

[0056] Writing real-time fundamental component parameters: ; In the formula, Indicates the first Real-time fundamental component parameters of each sampling point; Indicates the real-time fundamental frequency; Indicates the peak amplitude of the fundamental voltage wave; Indicates the peak amplitude of the fundamental current wave; This indicates the phase offset of the fundamental voltage wave; This indicates the phase offset of the fundamental current wave.

[0057] Writing the real-time fundamental component parameter sequence corresponding to the current analysis window: ; In the formula, This represents the real-time fundamental component parameter sequence corresponding to the current analysis window; Indicates the first Real-time fundamental component parameters of each sampling point; Indicates the sampling point number; This indicates the total number of synchronous sampling points within the current analysis window.

[0058] For example, taking the original electrical signal generated in step S1, the low-pass filter smoothing coefficient is taken as... The rated fundamental frequency is taken as The synchronous sampling period is taken .

[0059] exist At that time, the in-phase component of the output voltage of the quadrature signal generator Voltage quadrature components Current in-phase component Orthogonal components of current Fundamental wave tracking phase take .

[0060] Will , and Substituting into the formula for the direct-axis component of voltage, we get: ;Will , and Substituting into the formula for the cross-axis component of voltage, we get: ;Will , and Substituting into the formula for the direct-axis component of the current, we get: ;Will , and Substituting into the formula for the quadrature-axis components of the current, we get: The voltage smoothing direct-axis component at the 1599th sampling point is... The voltage smoothing cross-axis component is The smoothed direct-axis component of the current is The smoothed cross-axis component of the current is .

[0061] Will , and Substituting into the formula for the direct-axis component of voltage smoothing, we get: ;Will , and Substituting into the formula for smoothing the cross-axis components of voltage, we get: ;Will , and Substituting into the formula for smoothing the direct-axis components of the current, we get: ;Will , and Substituting into the formula for smoothing the cross-axis components of the current, we get: ;Will and Substituting into the normalized phase error formula, we get: The phase error integral state at the 1599th sampling point is taken as follows: ,Will , and Substituting into the integral state formula for phase error, we get: The proportionality coefficient is taken as follows: The integral coefficient is taken as ,Will , , and Substituting into the real-time frequency correction formula, we get: ;Will and Substituting into the real-time fundamental frequency formula, we get: ;Will , and Substituting into the fundamental wave tracking phase formula, we get: ;Will and Substituting into the formula for the peak amplitude of the fundamental voltage wave, we get: ;Will and Substituting into the formula for the peak amplitude of the fundamental current wave, we get: ;Will and Substituting into the formula for the phase shift of the fundamental voltage, we get: ;Will and Substituting into the formula for the phase shift of the fundamental current, we get: Therefore, the real-time fundamental component parameters of the 1600th sampling point are: The current analysis window contains For each synchronous sampling point, real-time fundamental component parameters are generated using the same processing method, resulting in a real-time fundamental component parameter sequence: Low-pass filtering converts the voltage quadrature axis component from... Smoothing The quadrature axis component of the current is changed from Smoothing This reduces the impact of transient disturbances on the estimation of fundamental wave parameters.

[0062] The real-time fundamental frequency was determined to be The peak amplitude of the fundamental voltage wave was determined to be The peak amplitude of the fundamental current wave was determined to be .

[0063] Real-time fundamental component parameter sequence Output to step S3 to generate a fundamental component reconstruction signal that matches the dynamic changes of the fundamental wave in the actual power grid, point by point.

[0064] S3. Based on the real-time fundamental component parameters, perform digital synthesis to generate a fundamental component reconstructed signal that matches the fundamental component in the original electrical signal; In one embodiment of the present invention, step S3 includes the following steps: Read the real-time fundamental frequency, voltage fundamental peak amplitude, current fundamental peak amplitude, voltage fundamental phase offset, and current fundamental phase offset from the real-time fundamental component parameters; The cumulative phase angle of the digital oscillator is calculated based on the real-time fundamental frequency. The fundamental component reconstructed signal is generated based on the cumulative phase angle, the voltage fundamental peak amplitude, the current fundamental peak amplitude, the voltage fundamental phase offset, and the current fundamental phase offset.

[0065] Specifically, such as Figure 2 As shown, the real-time fundamental component parameters are input to the digital oscillator. The digital oscillator recursively accumulates the phase angle based on the real-time fundamental frequency corresponding to each sampling point, and synthesizes the voltage fundamental reconstructed value and the current fundamental reconstructed value respectively. The digital synthesis process only synthesizes the fundamental component and does not synthesize harmonics, pulse disturbances, communication disturbances, or broadband noise.

[0066] Cumulative phase angle writing: ; In the formula, Indicates the digital oscillator at the 1st The cumulative phase angle of each sampling point, in radians; Indicates the first The cumulative phase angle of each sampling point; Represents pi; Indicates the first Real-time fundamental frequency of each sampling point; Indicates the sampling period.

[0067] Writing the fundamental voltage reconstructed value: ; In the formula, Indicates the first The reconstructed voltage fundamental value at each sampling point, in volts; Indicates the peak amplitude of the fundamental voltage wave; Indicates the cumulative phase angle; This indicates the phase offset of the fundamental voltage wave; This represents the sine function.

[0068] Writing the fundamental reconstructed value of the current: ; In the formula, Indicates the first The reconstructed fundamental current value at each sampling point, in amperes; Indicates the peak amplitude of the fundamental current wave; Indicates the cumulative phase angle; This indicates the phase offset of the fundamental current wave.

[0069] No. Writing of fundamental wave reconstruction records: ; In the formula, Indicates the first One fundamental wave reconstruction record; Indicates the first Each synchronous sampling moment; This represents the reconstructed value of the fundamental voltage wave; This represents the reconstructed value of the fundamental current.

[0070] Writing the fundamental component reconstructed signal: ; In the formula, This represents the reconstructed fundamental component signal corresponding to the current analysis window; Indicates the first One fundamental wave reconstruction record; Indicates the sampling point number; This indicates the total number of sampling points in the current analysis window.

[0071] The real-time fundamental component parameters are updated according to a preset update cycle. When there is an amplitude or phase change between two adjacent sets of real-time fundamental component parameters, linear interpolation is performed between the two parameter update times to make the voltage fundamental reconstructed value and the current fundamental reconstructed value change continuously, avoiding the introduction of step components into the fundamental component reconstructed signal by discrete parameter updates.

[0072] For example, the real-time fundamental component parameters obtained in step S2 The cumulative phase angle is taken at the 1600th sampling point. .

[0073] The reconstructed value of the voltage fundamental frequency is: The reconstructed value of the fundamental current is: The 1600th fundamental frequency reconstruction record is as follows: The cumulative phase angle of the 1601st sampling point is: Following the same method, digital synthesis was performed sequentially on the 12,800 sampling points within the current analysis window to obtain: The reconstructed signal frequency of the fundamental component is 50.0186Hz, instead of a fixed 50Hz, while retaining the voltage fundamental peak amplitude of 311.000V and the current fundamental peak amplitude of 14.140A.

[0074] Therefore, the fundamental component reconstructed signal can follow the changes in the actual power grid fundamental frequency, providing a fundamental frequency reference with a high degree of matching for subsequent point-by-point differential analysis.

[0075] S4. Perform a differential operation on the original electrical signal and the reconstructed signal of the fundamental component to obtain a differential residual signal that highlights the non-fundamental component; In one embodiment of the present invention, step S4 includes the following steps: The original electrical signal and the fundamental component reconstructed signal are matched point by point according to a unified timestamp; Calculate the difference between the voltage sample value and the reconstructed voltage fundamental value, and the difference between the current sample value and the reconstructed current fundamental value, respectively. The difference is normalized using the corresponding fundamental peak amplitude to generate the differential residual signal.

[0076] Specifically, such as Figure 2 As shown, the original electrical signal and the reconstructed signal of the fundamental component use the same synchronous sampling time and the same sampling point number, so point-by-point differential can be performed directly. Point-by-point differential strips the real-time changing fundamental component from the original electrical signal, so that the harmonics, transient pulses, narrowband disturbances, frequency drift disturbances and broadband noise in the original electrical signal are retained in the differential result.

[0077] Voltage residual writing: ; In the formula, Indicates the first The voltage residual at each sampling point, in volts; This represents the voltage sample value in the original electrical signal; This represents the reconstructed value of the voltage fundamental frequency.

[0078] Current residual writing: ; In the formula, Indicates the first The current residual at each sampling point is expressed in amperes. This represents the current sample value in the original electrical signal; This represents the reconstructed value of the fundamental current.

[0079] Normalized voltage residual writing: ; In the formula, Indicates the first Normalized voltage residuals at each sampling point; Indicates voltage residual; This represents the peak amplitude of the fundamental voltage wave; the normalized voltage residual is a dimensionless value.

[0080] Normalized current residual writing: ; In the formula, Indicates the first Normalized current residuals at each sampling point; Indicates current residual; This represents the peak amplitude of the fundamental current wave; the normalized current residual is a dimensionless value.

[0081] No. Writing individual difference residual records: ; In the formula, Indicates the first One difference residual record; Indicates the synchronous sampling time; Indicates voltage residual; Indicates the current residual; This represents the normalized voltage residual; This represents the normalized current residual.

[0082] Differential residual signal writing: ; In the formula, This represents the differential residual signal corresponding to the current analysis window; Indicates the first One difference residual record; Indicates the sampling point number; This indicates the total number of sampling points in the current analysis window.

[0083] The voltage and current residuals retain their original physical dimensions and are used for subsequent interference waveform reconstruction and measurement data correction; the normalized voltage and current residuals eliminate the amplitude and dimensional differences between the voltage and current channels and are used for subsequent joint time-frequency energy calculations.

[0084] For example, the synchronous sampling record obtained in step S1 and the fundamental reconstructed record obtained in step S3 .

[0085] The voltage residual is: The current residual is: ; using the results obtained in step S2 Calculate the normalized voltage residual: ; using the results obtained in step S2 Calculate the normalized current residual: The 1600th difference residual record is as follows: Following the same point-by-point differencing and normalization rules, all synchronously sampled records within the current analysis window are processed. and the corresponding fundamental reconstructed record The data is processed to obtain the differential residual records. .

[0086] For example, the synchronous sampling time of the 1599th sampling point is After point-by-point differencing and normalization, the following is obtained: The synchronous sampling time of the 1601st sampling point is After point-by-point differencing and normalization, the following is obtained: The current analysis window length is 1 second, and the sampling frequency is [missing value]. Therefore, the total number of sampling points is .

[0087] All differential residual records are collected sequentially according to the sampling point number to obtain the differential residual signal: The 310.984V fundamental component in the original voltage value was stripped, and the remaining 5.216V voltage residual consisted of harmonics and interference components.

[0088] Comparing adjacent sampling points reveals that the normalized voltage residual suddenly increased from 0.00141 to 0.01677, and the normalized current residual suddenly increased from 0.00184 to 0.02963, indicating that... A significant non-fundamental transient disturbance occurred nearby. The differential residual signal consists of all differential residual records. The output is sent to step S5 for subsequent frame windowing and time-frequency decomposition.

[0089] S5. Perform time-frequency decomposition on the differential residual signal to generate a two-dimensional time-frequency perturbation spectrum; In one embodiment of the present invention, step S5 includes the following steps: The normalized voltage residual and normalized current residual in the differential residual signal are framed and windowed. Perform time-frequency transformation on each frame to obtain the time-frequency coefficients of the voltage residual and the time-frequency coefficients of the current residual; The time-frequency disturbance spectrum is generated by fusing the voltage residual time-frequency coefficient and the current residual time-frequency coefficient, and subtracting the normal operating baseline energy.

[0090] Specifically, such as Figure 3 As shown, the differential residual signal is arranged according to a length of The analysis window and length are The frame shift is used for framing, with adjacent analysis frames maintaining overlap. The analysis window uses a Hanning window, Hamming window, or Gaussian window. For each frame, a short-time Fourier transform is performed on the normalized voltage residual and the normalized current residual, resulting in time-frequency coefficients with joint distribution in both the time and frequency dimensions.

[0091] Analysis of the total number of frames written as: ; In the formula, This indicates the total number of analysis frames in the current analysis window; This represents the total number of sampling points for the differential residual signal; Indicates the number of sampling points contained in a single analysis window; This indicates the number of sampling points between the start positions of two adjacent analysis frames; This indicates the floor function.

[0092] The voltage residual time-frequency coefficient is written as follows: ; In the formula, Indicates the first The analysis frame, the first The voltage residual time-frequency coefficients corresponding to each frequency unit; Indicates the analysis frame number, and ; Indicates the frequency unit number; Indicates the sampling point number within the analysis window; This represents the normalized voltage residual at the corresponding sampling location; The analysis window indicates the first... Window function coefficients at each sampling location; Represents the base of the natural exponential function; Represents the imaginary unit; This represents the number of points in the Discrete Fourier Transform. It represents pi (π).

[0093] The time-frequency coefficient of the current residual is written as: ; In the formula, Indicates the first The analysis frame, the first The current residual time-frequency coefficients corresponding to each frequency unit; This represents the normalized current residual at the corresponding sampling location; the meanings of the other symbols are the same as those of the corresponding symbols in the formula for the voltage residual time-frequency coefficient.

[0094] No. The center frequency of a frequency unit is written as: ; In the formula, Indicates the first The center frequency of each frequency unit is expressed in Hertz. Indicates the frequency unit number; Indicates the sampling frequency; This represents the number of points in the Discrete Fourier Transform.

[0095] Joint time-frequency energy writing: ; In the formula, Indicates the first The analysis frame, the first The joint time-frequency energy corresponding to each frequency unit; The complex modulus of the time-frequency coefficient of the voltage residual; The complex modulus of the time-frequency coefficient of the current residual; The energy weights of the current channels are represented and taken after normalization. .

[0096] No. Normal operating baseline energy writing for each frequency unit: ; In the formula, Indicates the first The normal operating baseline energy of each frequency unit; Indicates the first The first normal operation history window in the The analysis frame, the first The combined time-frequency energy of each frequency unit; This indicates the sequence number of the historical window during normal operation; This represents the total number of historically operating windows that participated in the baseline statistics; This indicates median operations.

[0097] Writing the perturbation energy in the time-frequency perturbation spectrum: ; In the formula, Indicates the first The analysis frame, the first The disturbance energy corresponding to each frequency unit; Represents the combined time-frequency energy; Indicates the baseline energy for normal operation; This indicates that the operation takes the larger value; the lower limit of the value is 0 to eliminate random fluctuations below the normal baseline.

[0098] Writing a two-dimensional time-frequency perturbation graph: ; In the formula, Represents a two-dimensional time-frequency perturbation spectrum; This represents the energy of the disturbance at a single time-frequency location. Indicates the sequence number of the analysis frame; Indicates the frequency unit number within the non-negative frequency range; Indicates the total number of analyzed frames; This represents the number of points in the Discrete Fourier Transform.

[0099] The normal operating baseline energy is used to eliminate inherent noise, stable harmonics, and sampling quantization errors in the electricity meter. After baseline subtraction, the time-frequency disturbance spectrum mainly retains the disturbance energy that is newly emerging or significantly enhanced relative to the normal operating state.

[0100] For example, taking the differential residual signal obtained in step S4, the analysis window length is determined. Frame shift Discrete Fourier Transform Points sampling frequency Total number of sampling points .

[0101] The total number of frames analyzed is: The center frequency of the 12th frequency unit is: In the 20th analysis frame and the 12th frequency unit, we obtained: ; ;Pick The combined time-frequency energy is: The normal operating baseline energy of the 12th frequency unit is taken as follows: The disturbance energy is: Since the center frequency corresponding to the 12th frequency unit is ,therefore This represents the perturbation energy at the 300Hz center frequency position in the 20th analysis frame of the two-dimensional time-frequency perturbation spectrum.

[0102] The two-dimensional time-frequency perturbation spectrum can be represented in matrix form as follows: Calculation results show that the disturbance energy at 300 Hz is significantly higher than the normal operating baseline.

[0103] After the disturbance energy appears continuously in adjacent analysis frames and adjacent frequency units, it forms a local energy cluster, which transforms the short-term weak interference in the one-dimensional differential residual into a divisible structure on the two-dimensional time-frequency plane.

[0104] S6. Based on the time-frequency perturbation spectrum, extract the perturbation morphology features characterizing the spatial distribution of the perturbation energy and the perturbation evolution features characterizing the temporal variation of the perturbation energy; In one embodiment of the present invention, step S6 includes the following steps: Adaptive threshold segmentation and connected component analysis are performed on the time-frequency perturbation spectrum to identify and segment independent perturbation energy clusters; Calculate the geometric moments, time span, frequency span, elongation, and principal direction of each of the perturbation energy clusters to generate perturbation morphological features; Within a continuous time window, the perturbation energy clusters are correlated across windows to extract their occurrence period, duration, frequency drift rate, and energy fluctuation coefficient, thereby generating perturbation evolution characteristics.

[0105] Specifically, such as Figure 3 As shown, an adaptive segmentation threshold is first established based on the mean energy and standard deviation of the energy at each frequency unit of the normal operating baseline. Then, the time-frequency perturbation map is converted into a binary perturbation mask based on the adaptive segmentation threshold. Subsequently, eight-neighborhood connectivity analysis is used to identify mutually independent perturbation energy clusters.

[0106] No. The adaptive segmentation threshold for each frequency unit is written as follows: ; In the formula, Indicates the first Adaptive segmentation threshold for each frequency unit; This indicates the normal operation history window is in the [number]th [year]. Average disturbance energy at each frequency unit; This indicates that the normal operation history window is in the [number]th [period]. Standard deviation of disturbance energy on each frequency unit; This represents the threshold adjustment coefficient.

[0107] Binary perturbation mask writing: ; In the formula, Indicates the first The analysis frame, the first Binary perturbation flags corresponding to each frequency unit; This represents the perturbation energy in the time-frequency perturbation spectrum; Indicates the adaptive segmentation threshold; when the condition within the square brackets is met. When the conditions are not met .

[0108] No. The weighted geometric moments of a perturbation energy cluster are written as follows: ; In the formula, Indicates the first A perturbed energy cluster Order-weighted geometric moments; Indicates the first The set of time-frequency locations contained in a perturbation energy cluster; Indicates the sequence number of the analysis frame; Indicates the frequency unit number; Indicates the order of moments in the time dimension; Indicates the order of the frequency dimension moments; This represents the disturbance energy at the corresponding time and frequency position.

[0109] No. Writing about the time centroid of a perturbation energy cluster: ; In the formula, Indicates the first The time centroid of a perturbation energy cluster; Indicates the first The first-order time moment of a perturbation energy cluster; Indicates the first The total perturbation energy of each perturbation energy cluster.

[0110] No. The frequency centroid of a perturbation energy cluster is written as follows: ; In the formula, Indicates the first The frequency centroid of a perturbation energy cluster; Indicates the first The first-order frequency moment of a perturbation energy cluster; Indicates the first The total perturbation energy of each perturbation energy cluster.

[0111] No. Writing the temporal span of a perturbation energy cluster: ; In the formula, Indicates the first The time span of a perturbation energy cluster, in seconds; Indicates the first The largest analysis frame number contained in each perturbation energy cluster; Indicates the first The smallest analysis frame number contained in each perturbation energy cluster; Indicates frame shift; Indicates the sampling period.

[0112] No. The frequency domain span of a perturbation energy cluster is written as follows: ; In the formula, Indicates the first The frequency domain span of a perturbation energy cluster, in Hertz; Indicates the first The largest frequency unit number contained in a perturbation energy cluster; Indicates the first The smallest frequency unit number contained in each perturbation energy cluster; Indicates the sampling frequency; This represents the number of points in the Discrete Fourier Transform.

[0113] No. The expansion rate of a perturbation energy cluster is written as: ; In the formula, Indicates the first The extent of a perturbation energy cluster; Indicates the first The larger eigenvalues ​​of the covariance matrix of the perturbation energy clusters; Indicates the first The smaller eigenvalues ​​of the covariance matrix of the perturbation energy clusters; .

[0114] No. The main direction of writing for a perturbation energy cluster: ; In the formula, Indicates the first The main direction of the perturbation energy cluster; Indicates the first The second-order central moment of a perturbation energy cluster along the time dimension; Indicates the first The second-order central moment of a perturbation energy cluster along the frequency dimension; Indicates the first The time-frequency mixing central moment of a perturbation energy cluster; This represents the arctangent function in the four quadrants.

[0115] No. The morphological descriptor for a perturbation energy cluster is written as follows: ; In the formula, Indicates the first A morphological descriptor for a perturbation energy cluster; Indicates the first The number of effective time-frequency locations contained in each perturbation energy cluster; Indicates the time span; Indicates frequency domain span; Indicates elongation; Indicates the main direction; This represents the total disturbance energy.

[0116] Writing about the morphological characteristics of disturbances: ; In the formula, This indicates the perturbation morphology characteristics corresponding to the current analysis window; Indicates the first A morphological descriptor for a perturbation energy cluster; Indicates the cluster number of the perturbation energy; This indicates the total number of perturbation energy clusters within the current analysis window.

[0117] Between consecutive analysis windows, cross-window correlations are performed based on the temporal centroid, frequency centroid, temporal overlap, and frequency overlap of perturbation energy clusters. Perturbation energy clusters that meet the correlation conditions are connected in chronological order to form the perturbation evolution trajectory.

[0118] No. Writing the evolutionary trajectory of a perturbation: ; In the formula, Indicates the first The trajectory of the perturbation evolution; Indicates the first The first of the perturbation evolution trajectories A cluster of perturbational energy; Indicates the cluster index within the trajectory; Indicates the first The number of perturbation energy clusters contained in each perturbation evolution trajectory.

[0119] No. Writing the occurrence cycle of the evolutionary trajectory of the perturbation: ; In the formula, Indicates the first The average occurrence period of each perturbation evolution trajectory, in seconds; Indicates the first The first perturbation evolution trajectory The starting point of the next similar disturbance; Indicates the starting time of the previous occurrence of the same type of disturbance; This indicates the number of times the same type of disturbance occurs.

[0120] No. Writing the duration of a perturbation evolution trajectory: ; In the formula, Indicates the first The duration of a single perturbation evolution trajectory, in seconds; Indicates the start time of a single disturbance; This indicates the end time of a single disturbance.

[0121] No. The frequency drift rate of the perturbation evolution trajectory is written as: ; In the formula, Indicates the first The frequency drift rate of the perturbation evolution trajectory, in Hertz per second; Indicates the frequency centroid at the start of a single disturbance; The frequency centroid represents the point at which a single disturbance ends; Indicates the duration of a single event.

[0122] No. The energy fluctuation coefficient of the perturbation evolution trajectory is written as: ; In the formula, Indicates the first The energy fluctuation coefficient of the perturbation evolution trajectory; The standard deviation of the disturbance energy within the trajectory; This represents the average value of the disturbance energy within the trajectory.

[0123] No. Evolutionary descriptor writing for the perturbation trajectory: ; In the formula, Indicates the first Evolutionary descriptor for the perturbation trajectory; Indicates the average occurrence period; Indicates duration; Indicates frequency drift rate; This represents the energy fluctuation coefficient.

[0124] Writing about perturbation evolution characteristics: ; In the formula, This indicates the perturbation evolution characteristics corresponding to the current analysis window; Indicates the first Evolutionary descriptor for the perturbation trajectory; Indicates the sequence number of the perturbation evolution trajectory; This represents the total number of perturbation evolution trajectories.

[0125] For example, the two-dimensional time-frequency perturbation spectrum generated in step S5. Near the 12th frequency unit, the average normal operating disturbance energy is taken as... The standard deviation of the disturbance energy during normal operation is taken as The threshold adjustment coefficient is taken as .

[0126] Will , and Substituting into the adaptive segmentation threshold formula, we get: Step S5 yields the disturbance energy of the 20th analysis frame and the 12th frequency unit as follows: ,because ,therefore: Threshold determination was performed on each time frequency position in the two-dimensional time-frequency perturbation spectrum in the same manner, and the first perturbation energy cluster was obtained through eight-neighbor connectivity analysis.

[0127] To facilitate the demonstration of the numerical substitution process for the weighted geometric moments, the representative time-frequency positions contained in the first perturbation energy cluster are written as: The disturbance energies corresponding to each time frequency position are as follows: ; ; ; ;right , and Substituting the time frequency positions and corresponding disturbance energies into the weighted geometric moment formula, we obtain the zeroth-order weighted geometric moment: ;right , and The first-order time-weighted geometric moments are obtained: ;right , and The first-order frequency-weighted geometric moments are obtained: ;right , and The second-order time-weighted geometric moments are obtained: ;right , and The second-order frequency-weighted geometric moments are obtained: ;right , and The time-frequency mixed weighted geometric moments are obtained as follows: ;Will and Substituting into the time centroid formula, we get: ;Will and Substituting into the frequency-centroid formula, we get: The minimum analysis frame number for the first perturbation energy cluster is... The maximum analysis frame number is Frame shifting The sampling period is taken .

[0128] Substituting the numerical values ​​into the time-domain span formula, we get: The minimum frequency unit number of the first perturbation energy cluster is... The maximum frequency unit number is Sampling frequency Discrete Fourier Transform Points .

[0129] Substituting the numerical values ​​into the frequency domain span formula, we get: ;according to , , , , and Calculate the covariance matrix of the first perturbation energy cluster to obtain two eigenvalues. and .

[0130] Substituting the numerical values ​​into the elongation formula, we get: The second central moment of the first perturbation energy cluster is: , and .

[0131] Substituting the numerical values ​​into the principal direction formula, we get: The number of effective time-frequency locations contained in the first perturbation energy cluster is: ,Will , , , , and Substituting the morphological descriptor, we get: The total number of perturbation energy clusters identified within the current analysis window is taken as... Therefore, the perturbation morphological characteristics are written as: Substituting the first morphological descriptor, we get: The calculation results show that the first perturbation energy cluster has a significantly greater extension in the time direction than in the frequency direction, and its frequency domain span is only 50 Hz. Therefore, this perturbation energy cluster belongs to a narrowband linear perturbation rather than a broadband sheet perturbation.

[0132] Within the continuous analysis window, perturbation energy clusters with similar frequency centroids, continuous time intervals, and projection overlap rates satisfying the correlation conditions are correlated across windows. The first perturbation evolution trajectory contains four sequentially occurring perturbation energy clusters, with the first perturbation energy cluster being the aforementioned... ,Right now The first perturbation evolution trajectory is written as follows: ;in, , , , and These represent the four perturbation energy clusters that are sequentially associated in the first perturbation evolution trajectory.

[0133] The start times of the four perturbation energy clusters are as follows: ; ; ; ;Will , , , and Substituting into the periodicity formula, we get: The start time of a single disturbance in the first disturbance evolution trajectory is... The end time is .

[0134] Substituting the numerical values ​​into the duration formula, we get: The frequency centroid at the start of a single disturbance is The frequency centroid at the end is .

[0135] Substituting the numerical values ​​into the frequency drift rate formula, we get: The average perturbation energy of the first perturbation evolution trajectory is taken as... The standard deviation of the disturbance energy is taken as .

[0136] Substituting the numerical values ​​into the energy fluctuation coefficient formula, we get: ;Will , , and Substituting the evolutionary descriptor, we get: The total number of perturbation evolution trajectories obtained within the current analysis window is taken as... Therefore, the perturbation evolution characteristics are written as: ; Disturbance morphology characteristics This indicates that the disturbance has a narrow-band linear structure, and the disturbance evolution trajectory This indicates that similar perturbation energy clusters repeat with a period of 0.100 s, and the perturbation evolution characteristics... This indicates that a single disturbance lasts for 0.020s and the frequency drifts at a rate of 1250Hz / s.

[0137] Step S6 will ultimately determine the perturbation morphological characteristics. and perturbation evolution characteristics The results are output to step S7 to generate a composite perturbation fingerprint that combines spatial morphological information and temporal evolution information.

[0138] S7. Couple the perturbation morphology features with the perturbation evolution features to generate a composite perturbation fingerprint; In one embodiment of the present invention, step S7 includes the following steps: The morphological descriptors in the perturbation morphological features are vector-quantized and mapped to generate a morphological codebook and a morphological code sequence. The transition relationship between adjacent morphological codes is modeled based on the morphological code sequence and the perturbation evolution characteristics to generate a state transition matrix; The composite perturbation fingerprint is generated by jointly encoding the morphological code occurrence frequency, the state transition matrix, and the perturbation evolution features.

[0139] Specifically, such as Figure 4 As shown, the morphological codebook is generated by clustering morphological descriptors from known interference samples. Each cluster center corresponds to a typical time-frequency morphology, including point-like morphology, narrowband line-like morphology, broadband patch-like morphology, and drift line-like morphology. Each morphological descriptor in the current analysis window is compared with all morphological prototypes in the morphological codebook in terms of distance and mapped to the morphological prototype with the smallest distance.

[0140] Morphological codebook writing: ; In the formula, Represents the morphological codebook; Indicates the first A prototype form; Indicates the prototype number; This indicates the total number of morphological prototypes contained in the morphological codebook.

[0141] No. morphological code writing for a morphological descriptor: ; In the formula, Indicates the first The morphological code corresponding to each morphological descriptor; Indicates the standardized number of A morphological descriptor; Indicates the first A prototype form; Indicates Euclidean distance; This represents the morphological prototype number that minimizes the Euclidean distance.

[0142] The morphological codes are arranged according to the occurrence time of the perturbation energy clusters to obtain the morphological code sequence: ; In the formula, This represents the morphological code sequence corresponding to the current analysis window; Indicates the first A morphological code for a morphological descriptor; Indicates the cluster number of the perturbation energy; This represents the total number of perturbation energy clusters.

[0143] No. The frequency of occurrence of each morphological code is written as follows: ; In the formula, Indicates the first The frequency of occurrence of each morphological code in the current analysis window; Indicates the first A morphological code; Indicates the total number of morphological codes; the value is 1 if the condition within the square brackets is true, and 0 if the condition is false.

[0144] Phenotype frequency vector writing: ; In the formula, Represents the morphological code frequency vector; Indicates the first The frequency of occurrence of each morphological code; Indicates the total number of morphological prototypes.

[0145] morphology code To morphological code State transition probability writing: ; In the formula, Represents morphological code To morphological code The state transition probability; Represents the morphic code in the morphic code sequence Immediately following morphological code The number of times it appears; Represents morphological code To morphological code The number of transfers; Indicates the transfer start form code; Indicates the transfer target morphology code; Indicates the summation sequence number of the transfer target morphological code; Indicates the total number of morphological prototypes; This represents a smoothing constant to prevent zero probability.

[0146] State transition matrix writing: ; In the formula, Represents the state transition matrix of the morphology code; Represents morphological code To morphological code The state transition probability; The state transition matrix contains lines and List.

[0147] Perturbation evolution feature pooling vector writing: ; In the formula, Represents the perturbation evolution feature pooling vector; This represents the average period of occurrence of all perturbation evolution trajectories; This represents the average duration of all perturbation evolution trajectories; This represents the average frequency drift rate of all perturbation evolution trajectories; This represents the average value of the energy fluctuation coefficient of the entire perturbation evolution trajectory.

[0148] Composite perturbation fingerprint writing: ; In the formula, This indicates the composite perturbation fingerprint corresponding to the current analysis window; Represents the morphological code frequency vector; This represents a one-dimensional vector obtained by expanding the state transition matrix in row order. This represents the pooling vector of perturbation evolution features.

[0149] The composite perturbation fingerprint simultaneously includes the morphological composition ratio of the interference in the time-frequency plane, the transfer law between different morphologies, and the dynamic parameters of the interference changing over time, thus mapping the static spatial structure and dynamic evolution law of the interference to the same fixed-length feature vector.

[0150] For example, the perturbation morphological features obtained in step S6 The morphological codebook contains There are morphological prototypes. Morphological code 1 corresponds to a dot-like morphology, morphological code 2 corresponds to a narrow-band line-like morphology, morphological code 3 corresponds to a wide-band sheet-like morphology, and morphological code 4 corresponds to a drifting line-like morphology.

[0151] The four prototype forms are as follows: ; ; ; Substituting the four morphological prototypes into the morphological codebook, we obtain: The first standardized morphological descriptor is: ;Will Compare with the four prototype forms respectively.

[0152] The Euclidean distance between the first normalized morphological descriptor and the first morphological prototype is: The Euclidean distance between the first normalized morphological descriptor and the second morphological prototype is: The Euclidean distance between the first normalized morphological descriptor and the third morphological prototype is: The Euclidean distance between the first normalized morphological descriptor and the fourth morphological prototype is: Substituting the four Euclidean distances into the morphological code calculation formula, we get: Therefore, the first standardized morphological descriptor is mapped to morphological code 2, corresponding to the narrowband linear morphology.

[0153] The other four normalized morphological descriptors are: ; ; ; Similarly, , , and Respectively with morphological codebook The four morphological prototypes in the model were compared using Euclidean distance, and the following results were obtained: ; ; ; Arrange the morphological codes sequentially according to the occurrence time of the five perturbation energy clusters to obtain the morphological code sequence: Substituting the morphological codes, we get: The frequencies of occurrence for each morphological code are as follows: ; ; ; The morphological code frequency vector is: The adjacent transitions in the morphological code sequence are respectively , , and ,therefore , , , All other transitions have a number of zeros. The smoothing constant is taken as... .

[0154] The transition probability from morphology code 2 to morphology code 1 is: The transition probability from morphology code 2 to morphology code 2 is: The transition probability from morphology code 2 to morphology code 3 is: The transition probability from morphology code 2 to morphology code 4 is: The transition probability from morphology code 4 to morphology code 1 is: The transition probability from morphology code 4 to morphology code 2 is: The transition probability from morphology code 4 to morphology code 3 is: The transition probability from morphology code 4 to morphology code 4 is: For morphology codes 1 and 3, which have no transition records, after smoothing, each transition probability is 0.25, and the state transition matrix is: Following the perturbation evolution features obtained in step S6, the pooling vector of the perturbation evolution features is: The composite perturbation fingerprint is: The morphological code frequency vector indicates that narrowband linear morphology and drifting linear morphology account for the majority, and the state transition matrix indicates that there is a stable round-trip transition between the two types of morphology. Combined with the occurrence period of 0.100s and the frequency drift rate of 1250Hz / s, a composite perturbation fingerprint that is distinct from fixed harmonics and random noise can be formed.

[0155] S8. Input the composite perturbation fingerprint into the interference identification model, and the interference identification model performs classification processing to obtain the interference type identification result.

[0156] In one embodiment of the present invention, step S8 includes the following steps: Obtain an interference sample library containing various known interference signal samples. Perform fundamental wave tracking, fundamental wave reconstruction, differential residual extraction, time-frequency decomposition, and composite perturbation fingerprint generation on each known interference signal sample to obtain a sample fingerprint set. A classifier is trained using the sample fingerprint set and the corresponding known interference type labels to generate an interference recognition model; The composite perturbation fingerprint is input into the interference recognition model to calculate the classification probability of each candidate interference type, thereby obtaining the interference type recognition result and classification confidence. Based on the interference type identification results, target anti-interference strategies are matched, and low-confidence composite perturbation fingerprints are clustered and updated online.

[0157] Specifically, such as Figure 4 As shown, the known interference types in the interference sample library include benign harmonics, PLC communication interference, wireless coupling interference, malicious pulse injection interference, and other labeled interference types. Each known interference signal sample is processed according to steps S1 to S7 to generate a corresponding sample composite perturbation fingerprint, and the sample composite perturbation fingerprint is combined with the known interference type label to form training data.

[0158] Writing the training dataset for the interference recognition model: ; In the formula, This represents the training dataset for the interference identification model; Indicates the first A composite perturbation fingerprint corresponding to a known interference signal sample; Indicates the first The known interference type label corresponding to each known interference signal sample; Indicates the sequence number of the known interference signal sample; This represents the total number of known interference signal samples in the training dataset.

[0159] The composite perturbation fingerprint is unfolded according to a preset feature order, and then normalized and feature mapped based on the statistical parameters of the training data to obtain the classification input vector.

[0160] Classification input vector writing: ; In the formula, This represents the classification input vector for the interference recognition model; This represents the processing function for unfolding, normalizing, and feature mapping of composite perturbation fingerprints; This indicates a composite perturbation fingerprint to be identified; Represents the first element in the classification input vector. Each feature component; Indicates the sequence number of the characteristic component; This represents the feature dimension of the classification input vector.

[0161] No. Writing category scores for each candidate interference type: ; In the formula, Indicates the first Category scores for each candidate interference type; Indicates the first The model weight vector corresponding to each candidate disturbance type; This represents the transpose of the model weight vector; Represents the classification input vector; Indicates the first Bias parameters corresponding to each candidate interference type; Indicates the candidate interference type number.

[0162] No. Writing the classification probability of each candidate interference type: ; In the formula, Indicates the first The classification probability of each candidate interference type; Indicates the first Category scores for each candidate interference type; Indicates the first Category scores for each candidate interference type; Indicates the summation index of the candidate interference type; Indicates the total number of candidate interference types; This represents the base of the natural exponential function.

[0163] Writing the Interference Type Identification Results: ; In the formula, This indicates the result of interference type identification; Indicates the first The classification probability of each candidate interference type; This represents the index of the candidate interference type that maximizes the classification probability. This indicates the total number of candidate interference types.

[0164] Writing Classification Confidence: ; In the formula, Indicates classification confidence level; Indicates the first The classification probability of each candidate interference type; This indicates the operation of finding the maximum value.

[0165] Writing the training loss of the interference recognition model: ; In the formula, This represents the training loss of the interference recognition model; Indicates the total number of training samples; Indicates the total number of candidate interference types; Indicates the first Does the training sample belong to the ? There are 1 candidate interference types; if a type is found, the value is 1; otherwise, the value is 0. The interference recognition model determines the first The training sample belongs to the first... The probability of each candidate interference type; Represents the natural logarithm; Represents the regularization coefficient; Indicates the first The model weight vector corresponding to each candidate disturbance type; This represents the square norm of the model weight vector.

[0166] When the classification confidence level is not lower than the confidence threshold, the target anti-interference strategy is obtained from the strategy library based on the interference type identification result. When the target anti-interference strategy is metrological correction, a target interference time-frequency mask is established based on the location of the perturbation energy clusters in the time-frequency perturbation map.

[0167] Target interference time-frequency mask writing: ; In the formula, Indicates the first The analysis frame, the first The target interference time-frequency mask value corresponding to each frequency unit; This represents the set of time-frequency positions occupied by the target interference in the time-frequency plane; hour ,otherwise .

[0168] Writing the time-frequency coefficients of target interference: ; In the formula, Indicates channel In the The analysis frame, the first The target interference time-frequency coefficients corresponding to each frequency unit; Indicates voltage channel or current channel ; Indicates channel The corresponding original time-frequency coefficients; This represents the time-frequency mask value for target interference.

[0169] Perform an inverse discrete Fourier transform on the target interference time-frequency coefficients of each analysis frame to obtain the target interference reconstruction sequence within the analysis frame.

[0170] No. Within the analysis frame, the first The target interference reconstruction value at each sampling location is written as follows: ; In the formula, Indicates channel In the Within the analysis frame, the first Target interference reconstruction value at each sampling location; Indicates the sampling position number within the analysis frame; This represents the number of points in the Discrete Fourier Transform. Represents the time-frequency coefficients of target interference; Indicates the frequency unit number; Represents the base of the natural exponential function; Represents the imaginary unit; It represents pi (π).

[0171] The target interference reconstructed values ​​within each analysis frame are windowed and superimposed according to the frame shift to obtain the complete target interference waveform.

[0172] Target interference waveform writing: ; In the formula, Indicates channel In the The reconstructed value of the target interference at each sampling point; Indicates the total number of analyzed frames; Indicates the sequence number of the analysis frame; Indicates the first The analysis frame in the ... The window function coefficients corresponding to each sampling point; Indicates frame shift; Indicates the first The target interference reconstruction value at the sampling position within the corresponding frame of each analysis frame; the denominator represents the energy normalization term of the overlapping window function.

[0173] Writing post-purification measurement data: ; In the formula, Indicates channel In the Post-purification measurement data from each sampling point; Indicates channel In the The original sampled values ​​of each sampling point; Indicates channel In the The target interference reconstruction value of each sampling point.

[0174] When the classification confidence is lower than the confidence threshold, the corresponding composite perturbation fingerprint is written into the unknown perturbation fingerprint pool.

[0175] Unknown perturbation fingerprint pool writing: ; In the formula, This indicates an unknown perturbation fingerprint pool; Indicates the first One composite perturbation fingerprint to be identified; Indicates the first Classification confidence of a composite perturbation fingerprint to be identified; Indicates the confidence threshold; This indicates the serial number of the composite perturbation fingerprint to be identified.

[0176] No. The size determination result of an unknown perturbation cluster is written as follows: ; In the formula, Indicates the first Does the clustering of unknown perturbations meet the model update conditions? Indicates the first The set of unknown perturbation fingerprints contained in an unknown perturbation cluster; Indicates the first The number of unknown perturbation fingerprints contained in each unknown perturbation cluster; Indicates the scale threshold; when the condition is met. This triggers the generation of novel interference labels and the retraining of the interference identification model; if the conditions are not met... .

[0177] For example, to include a voltage channel and current path Taking electricity data as an example, the first training sample is selected from the training dataset, i.e. In the original electrical signal corresponding to the first training sample, the 1600th sampling point is selected, i.e. This explains the process of interference classification, target interference reconstruction, and measurement correction.

[0178] The training dataset contains There are 10 known interference signal samples. The first sample is malicious pulse injection interference, the second sample is PLC communication interference, and the third sample is benign harmonics.

[0179] The training dataset is: The candidate interference types are arranged in the order of benign harmonics, PLC communication interference, and malicious pulse injection interference.

[0180] The first training sample belongs to malicious pulse injection interference, therefore its label is: Composite perturbation fingerprint of the first training sample After unfolding, normalizing, and feature mapping, the classification input vector is obtained: The model weight vector and bias parameters corresponding to the benign harmonic category are respectively taken as follows: ; ;Will , and Substituting into the category score formula, we get: The model weight vector and bias parameters corresponding to the PLC communication interference category are respectively taken as follows: ; ;Will , and Substituting into the category score formula, we get: The model weight vector and bias parameters corresponding to the malicious pulse injection interference category are respectively taken as follows: ; ;Will , and Substituting into the category score formula, we get: The category score vector is: The index values ​​corresponding to the scores of the three categories are as follows: ; ; The sum of the exponent values ​​is: The probability of benign harmonic classification is: The probability of PLC communication interference classification is: The probability of malicious pulse injection interference is: The interference type identification result is as follows: The classification confidence level is: The true label of the first training sample is Regularization term selection Then the loss corresponding to the first training sample is: ; Confidence threshold is taken ,because The interference type identification result of the first training sample was determined to be malicious pulse injection interference, and a joint strategy of metering correction and active alarm was matched.

[0181] The target interference is located in the 12th frequency cell and its conjugate frequency cell of the 23rd analysis frame. The time-frequency location set of the target interference includes... and .

[0182] therefore: ; For the voltage channel, the original time-frequency coefficients of the 23rd analysis frame and the 12th frequency unit are taken as follows: ;Will and Substituting into the formula for the time-frequency coefficients of target interference, we get: The corresponding time-frequency coefficients of the conjugate frequency unit are: For the current channel, the original time-frequency coefficients of the 23rd analysis frame and the 12th frequency unit are taken as follows: ;Will and Substituting into the formula for the time-frequency coefficients of the target interference, we get: The corresponding time-frequency coefficients of the conjugate frequency unit are: The sampling position of the 1600th sampling point within the 23rd analysis frame is: In the target interference time-frequency coefficients of the voltage channel, only the 12th frequency unit and the 500th conjugate frequency unit are retained.

[0183] Will , , and Substituting into the inverse discrete Fourier transform formula, we get: Similarly, and Substituting into the inverse discrete Fourier transform formula, we get: At the 1600th sampling point, the target interference is only reconstructed from the 23rd analysis frame, and after normalization and overlapping by a window function, the following is obtained: ; The original voltage value of the first training sample at the 1600th sampling point is The original current value is .

[0184] Substituting the original voltage value and the reconstructed voltage target interference value into the formula for the purified metering data, we get: Substituting the original current value and the reconstructed current target interference value into the purified metering data formula, we obtain: The purified voltage value of 311.02V is close to the voltage fundamental reconstructed value of 310.984V obtained in step S3, and the purified current value of 13.94A is close to the current fundamental reconstructed value of 13.931A obtained in step S3, indicating that the malicious pulse injection component has been deducted from the measurement data.

[0185] The main station system synchronously receives alarm information including malicious pulse injection, occurrence time of 0.100s to 0.120s, frequency range of 300Hz to 325Hz, and classification confidence level of 0.8451. For the fourth composite perturbation fingerprint to be identified... The classification confidence score output by the interference identification model is: ;because: Therefore, the fourth composite perturbation fingerprint to be identified is written into the unknown perturbation fingerprint pool: When with The number of composite perturbation fingerprints belonging to the same unknown perturbation cluster reached Scale threshold is taken At that time, we obtained: ;No. If an unknown perturbation cluster meets the model update condition, it is added to the training dataset after verification with a new type of perturbation label. This information is then used to retrain the interference identification model, thereby improving the ability of the energy meter to identify and adapt to new non-stationary interferences.

[0186] Based on the same inventive concept, such as Figure 5 As shown, the present invention also provides a time-frequency hybrid anti-interference identification system for electricity meters, the system comprising: The synchronous acquisition module is used to acquire voltage and current signals at the metering points of the electricity meter and generate raw electrical signals. The fundamental wave tracking module is used to track the fundamental wave component of the original electrical signal in real time, and obtain a set of real-time fundamental wave component parameters characterizing the dynamic changes of the fundamental wave. The fundamental wave reconstruction module is used to perform digital synthesis based on the real-time fundamental wave component parameters to generate a fundamental wave component reconstruction signal that matches the fundamental wave component in the original electrical signal. The residual extraction module is used to perform differential operation on the original electrical signal and the fundamental component reconstructed signal to obtain a differential residual signal that highlights the non-fundamental components; The time-frequency spectrum generation module is used to perform time-frequency decomposition on the differential residual signal to generate a two-dimensional time-frequency perturbation spectrum; The perturbation feature extraction module is used to extract perturbation morphology features characterizing the spatial distribution of perturbation energy and perturbation evolution features characterizing the temporal variation of perturbation energy based on the time-frequency perturbation map. A composite fingerprint generation module is used to couple the perturbation morphological features with the perturbation evolution features to generate a composite perturbation fingerprint; The interference identification module is used to classify the composite disturbance fingerprint input into the interference identification model to obtain the interference type identification result, and to perform metering correction, active alarm or online model update based on the interference type identification result.

[0187] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A time-frequency hybrid anti-interference identification method for electricity meters, characterized in that, The method includes: Collect voltage and current signals from the electricity meter's metering points to generate raw electrical signals; The original electrical signal is tracked in real time for its fundamental component to obtain a set of real-time fundamental component parameters characterizing the dynamic changes of the fundamental wave. Based on the real-time fundamental component parameters, digital synthesis is performed to generate a fundamental component reconstructed signal that matches the fundamental component in the original electrical signal; The original electrical signal and the reconstructed signal of the fundamental component are subjected to a differential operation to obtain a differential residual signal that highlights the non-fundamental components; The differential residual signal is decomposed into time and frequency to generate a two-dimensional time and frequency perturbation spectrum; Based on the time-frequency perturbation map, perturbation morphology features characterizing the spatial distribution of perturbation energy and perturbation evolution features characterizing the temporal variation of perturbation energy are extracted. The perturbation morphology features are coupled with the perturbation evolution features to generate a composite perturbation fingerprint; The composite perturbation fingerprint is input into the interference identification model, which then performs classification processing to obtain the interference type identification result.

2. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, The real-time tracking of the fundamental component of the original electrical signal to obtain a set of real-time fundamental component parameters characterizing the dynamic changes of the fundamental wave includes: The original electrical signal is subjected to orthogonal component generation and synchronous rotating coordinate system transformation to obtain a DC component containing fundamental wave information and an AC component containing harmonic and interference information. The DC component is subjected to low-pass filtering to obtain a smooth DC component with high-frequency fluctuations removed. The real-time frequency, amplitude, and phase of the original electrical signal are calculated based on the smoothed DC component, and the real-time fundamental component parameters are generated.

3. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, The perturbation morphology features extracted based on the time-frequency perturbation spectrum to characterize the spatial distribution of perturbation energy include: Adaptive threshold segmentation and connected component analysis are performed on the time-frequency perturbation spectrum to identify and segment independent perturbation energy clusters; Calculate the geometric moments, time span, frequency span, and principal direction of each of the perturbation energy clusters to obtain morphological descriptors characterizing the shape, size, and orientation; The morphological descriptors are combined to generate the perturbation morphological features.

4. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 3, characterized in that, The perturbation evolution features extracted based on the time-frequency perturbation spectrum to characterize the temporal variation of the perturbation energy include: The perturbation energy clusters are correlated across windows within a continuous time window to construct their motion trajectory; The dynamic parameters of the motion trajectory are extracted, including the cluster occurrence period, duration, frequency drift rate and energy fluctuation coefficient, to obtain the evolutionary descriptor; The evolutionary descriptors are combined to generate the perturbation evolutionary features.

5. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, The step of coupling the perturbation morphological features with the perturbation evolution features to generate a composite perturbation fingerprint includes: The morphological descriptors in the perturbation morphological features are vector-quantized and mapped to generate a morphological codebook and the corresponding morphological code sequence. The transition relationship between adjacent morphological codes is modeled based on the perturbation evolution characteristics and the morphological code sequence to generate a state transition matrix describing the transition rules between morphologies; The occurrence frequency of the morphological codebook, the state transition matrix, and the perturbation evolution features are jointly encoded to generate the composite perturbation fingerprint.

6. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, Before inputting the composite perturbation fingerprint into the interference recognition model, the method further includes generating the interference recognition model: Obtain an interference sample library containing a variety of known interference signal samples; For each interference signal sample in the interference sample library, perform real-time tracking of fundamental component, reconstruction of fundamental component, extraction of differential residual, time-frequency decomposition, extraction of perturbation features and generation of composite perturbation fingerprint in sequence to obtain a sample fingerprint set; The classifier is trained using the sample fingerprint set and its corresponding known interference type labels to generate the interference recognition model.

7. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, After obtaining the interference type identification result, the method further includes: Based on the interference type identification result, the target anti-interference strategy is matched and obtained from the strategy library that stores the mapping relationship between interference type and response strategy; If the target anti-interference strategy is metrological correction, then the interference waveform is reconstructed based on the disturbance energy clusters corresponding to the composite disturbance fingerprint in the time-frequency disturbance spectrum, and the interference waveform is subtracted from the original electrical signal to obtain the purified metrological data; If the target anti-interference strategy is active alarm, then alarm information including interference type, occurrence time, frequency range and classification confidence level is generated and reported to the main station system.

8. The time-frequency hybrid anti-interference identification method for an electricity meter according to claim 6, characterized in that, The method further includes online updating of the interference identification model: The interference recognition model is used to classify the composite perturbation fingerprint and obtain the classification confidence score; The composite perturbation fingerprint with a classification confidence level below the confidence threshold is identified as an unknown perturbation fingerprint; Online clustering analysis is performed on the unknown perturbation fingerprint. When the number of samples in any cluster reaches a scale threshold, the cluster is marked as a new type of interference label, and the interference sample library is updated and the interference recognition model is retrained using the new interference label.

9. The time-frequency hybrid anti-interference identification method for electricity meters according to claim 1, characterized in that, The process of collecting voltage and current signals from the electricity meter's metering points to generate raw electrical signals includes: Obtain the synchronous sampling clock generated by the high-precision clock source; The voltage channel and current channel are sampled synchronously using the synchronous sampling clock to obtain a synchronous voltage-current sequence containing the precise phase relationship between voltage and current. The synchronous voltage and current sequences are combined into a multi-dimensional time sequence according to a unified timestamp, which serves as the original electrical signal.

10. A time-frequency hybrid anti-interference identification system for electricity meters, characterized in that, The system includes: The synchronous acquisition module is used to acquire voltage and current signals at the metering points of the electricity meter and generate raw electrical signals. The fundamental wave tracking module is used to track the fundamental wave component of the original electrical signal in real time, and obtain a set of real-time fundamental wave component parameters characterizing the dynamic changes of the fundamental wave. The fundamental wave reconstruction module is used to perform digital synthesis based on the real-time fundamental wave component parameters to generate a fundamental wave component reconstruction signal that matches the fundamental wave component in the original electrical signal. The residual extraction module is used to perform differential operation on the original electrical signal and the fundamental component reconstructed signal to obtain a differential residual signal that highlights the non-fundamental components; The time-frequency spectrum generation module is used to perform time-frequency decomposition on the differential residual signal to generate a two-dimensional time-frequency perturbation spectrum; The perturbation feature extraction module is used to extract perturbation morphology features characterizing the spatial distribution of perturbation energy and perturbation evolution features characterizing the temporal variation of perturbation energy based on the time-frequency perturbation map. A composite fingerprint generation module is used to couple the perturbation morphological features with the perturbation evolution features to generate a composite perturbation fingerprint; The interference identification module is used to classify the composite disturbance fingerprint input into the interference identification model to obtain the interference type identification result, and to perform metering correction, active alarm or online model update based on the interference type identification result.