Method and System for Harmonic Multidimensional Component Separation in Electricity Metering

Abstract

This invention relates to the field of power monitoring, specifically to a method and system for separating multi-dimensional harmonic components in power metering. The method includes the following steps: synchronously acquiring three-phase voltage and current observation signals at a fixed frequency, and outputting standardized synchronous observation signals; fitting and stripping strong fundamental components using a recursive least squares method to obtain residual signals containing harmonics across the entire frequency band; performing a multi-scale variable window S-transform on the residual signals to achieve decoupling and sparse characterization of cross-modulation components, and constructing three-dimensional time-frequency and channel tensors; estimating the number of harmonic sources and identifying the mixing matrix based on tensor CP decomposition, and completing sparse reconstruction using orthogonal matching pursuit to restore the full-frequency time-domain signals of each harmonic source; verifying the reconstruction accuracy of the separation results, extracting harmonic features of each harmonic source, matching them with a typical power electronic disturbance source feature library to complete calibration, and outputting the final separation results. This invention is fully adaptable to actual metering conditions in the field and has excellent engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of power monitoring, specifically relating to a method and system for separating multidimensional harmonic components in power metering. Background Technology

[0002] Traditional harmonic separation is generally based on the positive definite / overdetermined ideal assumption, requiring the number of observation channels to be greater than or equal to the number of harmonic sources. However, actual power metering terminals can only provide six observation channels for three-phase voltage and current. In contrast, current new power systems often have dozens of power electronic disturbance sources such as photovoltaic inverters, charging piles, and energy storage converters at the same metering point, which is a typical underdetermined separation scenario. Traditional methods cannot adapt to this actual working condition.

[0003] The harmonic components of different disturbance sources in the field have cross-dimensional modulation problems, including sideband modulation of the fundamental and interharmonics, and nonlinear coupling of integer harmonics and subharmonics. Traditional blind source separation, S-transform, wavelet transform, etc., will have severe component aliasing under the dual conditions of underdetermined and strong coupling. Weak multidimensional components such as zero-sequence harmonics, higher harmonics, and sub / interharmonics with amplitudes much lower than the fundamental wave will be completely submerged by the strong fundamental wave component, and cannot achieve effective separation and accurate extraction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for separating harmonic multidimensional components in power metering, and at the same time, to provide a system for separating harmonic multidimensional components in power metering, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for separating harmonic multidimensional components in electricity metering includes the following steps: Three-phase voltage and current observation signals are synchronously acquired at a fixed frequency. Power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization are completed, and standardized synchronous observation signals are output. The recursive least squares method was used to fit and remove the strong fundamental wave component to obtain the residual signal containing harmonics across the entire frequency band. Perform a multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components, and construct a three-dimensional time-frequency and channel tensor; Based on tensor CP decomposition, the number of harmonic sources is estimated and the mixing matrix is ​​identified. Orthogonal matching pursuit is used to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. Verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

[0006] Furthermore, the process of completing power frequency synchronization alignment, outlier removal, DC bias removal, and amplitude normalization to output a standardized synchronous observation signal includes: using a digital phase-locked loop to perform power frequency period synchronization alignment based on zero-crossing detection for the acquired multi-channel observation signals; using a corresponding criterion to remove pulse-type outliers from the signals; using linear interpolation of adjacent effective sampling points to complete the corresponding missing data; using a moving average filter to remove the DC bias component from the signals; performing amplitude normalization on the signals that have undergone the aforementioned processing; and finally outputting a standardized multi-channel synchronous observation signal.

[0007] Furthermore, the step of fitting and stripping the strong fundamental wave component using the recursive least squares method to obtain the residual signal containing full-band harmonics includes: using the synchronous phase-locked power frequency phase corresponding to the standardized synchronous observation signal as a reference, constructing a time-domain fitting model containing the positive sequence, negative sequence, and zero sequence full components of the fundamental wave; using the recursive least squares algorithm to perform real-time fitting of the full components of the fundamental wave on multiple standardized synchronous observation signals respectively; subtracting the corresponding fitted full components of the fundamental wave from each original standardized synchronous observation signal to obtain the residual observation signal after fundamental wave stripping; the residual signal completely retains the full-band harmonic components in the original signal.

[0008] Furthermore, the step of performing a multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components includes: performing time-frequency domain transformation on the multi-path residual observation signals output from the previous step using a multi-scale variable window S-transform; adaptively matching Gaussian windows of corresponding window lengths for transformation processing on subharmonics, interharmonics, and integer higher harmonics components in different frequency bands; and achieving time-frequency domain sparse representation of harmonic components across the entire frequency band through adaptive adjustment of the Gaussian window width. This enables harmonic components of different frequencies and modulation types to form a unique single-peak sparse representation on the corresponding time-frequency scale, eliminating component aliasing caused by cross-dimensional harmonic modulation.

[0009] Furthermore, the construction of the three-dimensional time-frequency and channel tensor includes: performing dimensional consistency verification and alignment on the time-frequency sparse representation matrix output by the residual observation signal after multi-scale variable window S-transformation to ensure that the time scale dimension and frequency scale dimension of the corresponding matrix of each signal are completely matched; then, taking the observation channel as the third dimension, the dimensionally aligned time-frequency sparse representation matrix is ​​sequentially spliced ​​along the channel dimension to construct a three-dimensional time-frequency and channel tensor model, which fully preserves the full-dimensional feature information of the residual signal and serves as standardized input data for subsequent harmonic separation processing.

[0010] Furthermore, the estimation of the number of harmonic sources and identification of the mixing matrix based on tensor CP decomposition includes: for the constructed three-dimensional time-frequency and channel tensors, firstly, the tensor is subjected to low-rank approximation by minimizing the nuclear norm to suppress the aliasing components of time-frequency domain noise and sparse representation residues, and fully retain the main characteristic components of effective harmonic sources. Then, based on the three-dimensional tensor after low-rank approximation, the optimal rank of the tensor is determined by the Bayesian information criterion to obtain the actual number of harmonic sources at the measurement point. Subsequently, the tensor is subjected to CP decomposition by high-order orthogonal iteration to decompose it into a linear superposition of rank-1 tensors matching the number of harmonic sources. Finally, the factor matrix of the observation channel dimension is extracted based on the decomposition result as the mixing matrix, providing the core input for subsequent sparse reconstruction and time-domain separation of single harmonic source components.

[0011] Furthermore, the step of using orthogonal matching pursuit to complete sparse reconstruction and restore the full-band time-domain signal of each harmonic source includes: using the hybrid matrix identified in the previous step as the dictionary matrix for sparse reconstruction, using the time-frequency sparse characterization matrix corresponding to the residual observation signal output in the previous step as the reconstruction target, performing sparse solution point by point using orthogonal matching pursuit to obtain the time-frequency domain sparse coefficient matrix corresponding to each harmonic source, performing an inverse transformation on the coefficient matrix that matches the previous multi-scale variable window S-transform to restore the residual time-domain components of each harmonic source, and then allocating the fundamental wave full component according to the proportion of the fundamental wave transmission characteristics of each harmonic source carried by the hybrid matrix and superimposing them to finally obtain the complete time-domain signal of each harmonic source containing the fundamental wave and the full-band harmonic components.

[0012] Furthermore, the verification of the reconstruction accuracy of the separation result includes: acquiring the complete time-domain signals of each harmonic source obtained from the previous steps, linearly superimposing the identified hybrid matrix to obtain the reconstructed observation signal, calculating the error between the reconstructed observation signal and the original standardized synchronous observation signal output from the previous steps, comparing the error with a preset threshold, and determining that the reconstruction accuracy of the separation result meets the threshold requirement and the result is valid if the error does not meet the threshold requirement. Otherwise, returning to the corresponding previous steps for reprocessing is performed until the error meets the preset accuracy requirement.

[0013] Furthermore, the step of extracting harmonic features of each harmonic source and matching them with a feature library of typical power electronic disturbance sources to complete the calibration includes: performing spectral analysis on the complete time-domain signals of each separated harmonic source, extracting the feature parameters of the fundamental wave and the full-band harmonics, generating the harmonic feature vectors of the corresponding harmonic sources, constructing a feature library of inherent harmonic emission of typical power electronic disturbance sources, pre-storing the inherent feature parameters of the full-band harmonics of various disturbance sources, using a similarity calculation criterion to match the harmonic feature vectors of each harmonic source with the feature library parameters, completing the one-to-one calibration of the harmonic sources and the corresponding disturbance sources, and retaining the relevant calibration data.

[0014] Furthermore, the final output separation result includes: integrating the complete time-domain signals of each harmonic source obtained from the previous steps, the fundamental and full-band harmonic characteristic parameters, the calibration results and corresponding matching information of typical power electronic disturbance sources, recording the validity identifier and corresponding error value of this harmonic separation result, organizing it into a harmonic separation and disturbance source calibration report according to a standardized data format, and outputting it to the storage module and data interaction interface of the power metering terminal.

[0015] This application also discloses an electronic device, including: At least one processor; and

[0016] A memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the above-described method for separating harmonic components in power metering according to the present invention.

[0018] This application also discloses a harmonic multi-dimensional component separation system for electricity metering, including: The signal preprocessing module is used to synchronously acquire three-phase voltage and current observation signals at a fixed frequency, complete power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization, and output standardized synchronous observation signals. The fundamental wave stripping module is used to fit and strip the strong fundamental wave component using the recursive least squares method to obtain a residual signal containing harmonics across the entire frequency band. The time-frequency tensor construction module is used to perform multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components and construct three-dimensional time-frequency and channel tensors. The harmonic source separation module is used to estimate the number of harmonic sources and identify the mixing matrix based on tensor CP decomposition. It then uses orthogonal matching pursuit to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. The result verification and calibration module is used to verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

[0019] Beneficial effects: This application addresses the core pain points of underdetermined separation, strong harmonic coupling and aliasing, and weak components being submerged by strong fundamental waves in the metering of new power systems. It breaks through the limitations of the positive definite / overdetermined ideal assumptions of traditional harmonic separation. Through a time-progressive technical link, it achieves accurate separation of harmonic components across the entire frequency band from dozens of harmonic sources under 6 observation channels. It can effectively decouple cross-modulated harmonic components, accurately extract weak multidimensional components such as zero-sequence, higher-order, and second / inter-harmonics, and simultaneously complete the harmonic characteristic calibration of each power electronic disturbance source. The separation results are highly accurate and traceable, fully adaptable to actual metering conditions on site, and have excellent engineering application value. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the harmonic multi-dimensional component separation method for power metering of the present invention; Figure 2 is a flowchart of the implementation steps of signal preprocessing and fundamental wave stripping in the power metering harmonic multidimensional component separation method of the present invention; Figure 3 is a flowchart of the implementation steps of multi-scale variable window S-transform and three-dimensional time-frequency and channel tensor construction in the power metering harmonic multi-dimensional component separation method of the present invention; Figure 4 is a flowchart of the implementation steps of tensor CP decomposition and orthogonal matching tracking sparse reconstruction in the electric energy metering harmonic multidimensional component separation method of the present invention. Figure 5 is a flowchart of the implementation of the separation result reconstruction accuracy verification step in the electric energy metering harmonic multi-dimensional component separation method of the present invention; Figure 6 This is a comparison diagram of the time-domain waveforms and reconstruction errors of the signals before and after separation in the harmonic multi-dimensional component separation method for power metering of the present invention; Figure 7 This is a comparison diagram of the full-band harmonic amplitude spectrum before and after separation in the electric energy metering harmonic multi-dimensional component separation method of the present invention. Detailed Implementation

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

[0022] This invention provides a method for separating multidimensional harmonic components in electricity metering, such as... Figure 1 As shown, the steps include: Three-phase voltage and current observation signals are synchronously acquired at a fixed frequency. Power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization are completed, and standardized synchronous observation signals are output. The recursive least squares method was used to fit and remove the strong fundamental wave component to obtain the residual signal containing harmonics across the entire frequency band. Perform a multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components, and construct a three-dimensional time-frequency and channel tensor; Based on tensor CP decomposition, the number of harmonic sources is estimated and the mixing matrix is ​​identified. Orthogonal matching pursuit is used to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. Verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

[0023] The method of synchronously acquiring three-phase voltage and current observation signals at a fixed frequency, such as Figure 2 As shown, the implementation specifically includes: using the high-stability synchronous sampling clock of the power metering terminal as a reference, synchronously collecting six observation signals at the same metering point: three-phase voltage signals ua(t), ub(t), uc(t) and three-phase current signals ia(t), ib(t), ic(t). The sampling frequency is fixed at 12.8kHz, and the number of sampling points per power frequency cycle is 256.

[0024] The process of completing power frequency synchronization alignment, outlier removal, DC bias removal, and amplitude normalization outputs a standardized synchronization observation signal, such as... Figure 2 As shown, the implementation specifically includes: using a digital phase-locked loop based on zero-crossing detection to perform power frequency cycle synchronization alignment on the acquired 6 observation signals, controlling the phase alignment error within 1μs; using the 3σ criterion to remove pulse-type outliers from the 6 observation signals, and using linear interpolation of adjacent effective sampling points to complete the corresponding missing data; using a moving average filter with a fixed sliding window width of 2 power frequency cycles to remove the DC bias component from the 6 observation signals; performing amplitude normalization processing on the 6 observation signals after the above processing, linearly mapping the amplitude of each signal to the [-1,1] interval, eliminating the interference of amplitude differences between different channels on subsequent processing, and finally outputting standardized 6-channel synchronous observation signals.

[0025] The recursive least squares method is used to fit and remove the strong fundamental wave component, resulting in a residual signal containing harmonics across the entire frequency band. Figure 2 As shown, the specific implementation includes: using the synchronous phase-locked loop power frequency phase corresponding to the standardized synchronous observation signal as a reference, constructing a time-domain fitting model including the fundamental positive sequence, negative sequence, and zero sequence full components; and using a recursive least squares algorithm with a fixed forgetting factor of 0.995 to perform real-time fitting of the fundamental full components of the six standardized synchronous observation signals, with the fitting convergence accuracy set to... Subtract the corresponding fitted fundamental component from each original standardized synchronous observation signal to obtain 6 residual observation signals after fundamental stripping. The residual signals completely retain all subharmonic, interharmonic, and integer harmonic components in the original signal.

[0026] The multi-scale variable window S-transform is performed on the residual signal to achieve decoupling and sparse representation of the cross-modulation components, such as... Figure 3 As shown, the implementation specifically includes: for the 6 residual observation signals output from the previous steps, a multi-scale variable window S-transform is used for time-frequency domain transformation. For subharmonic components with frequencies below 50Hz and interharmonic components with frequencies in the range of 50Hz to 100Hz, a Gaussian long time window with a window length of ≥1 power frequency cycle is used for transformation processing to control the frequency resolution within 1Hz; for integer higher harmonic components with frequencies ≥100Hz, a Gaussian short time window with a window length of ≤1 / 2 power frequency cycle is used for transformation processing to control the time resolution within 1ms; through adaptive adjustment of the Gaussian window width, a sparse time-frequency representation of harmonic components across the entire frequency band is achieved, so that harmonic components of different frequencies and modulation types form a unique single-peak sparse representation on the corresponding time-frequency scale, completely eliminating component aliasing caused by sideband modulation of the fundamental wave and interharmonics, and nonlinear coupling between integer harmonics and subharmonics.

[0027] The construction of the three-dimensional time-frequency and channel tensor, such as Figure 3 As shown, the implementation specifically includes: first, performing dimension consistency verification and alignment on the time-frequency sparse representation matrices output from the multi-scale variable window S-transform of the 6 residual observation signals to ensure that the time-frequency sparse representation matrix corresponding to each signal has a fully matched time scale dimension and frequency scale dimension. The length of the time scale dimension strictly corresponds to the total number of sampling points within the analysis time, and the length of the frequency scale dimension fully covers the entire frequency range from the second harmonic to the highest analysis harmonic. Then, taking the observation channel as the third dimension, the 6 time-frequency sparse representation matrices that have completed dimension alignment are arranged with the three-phase voltage ua(t), ub(t), and uc(t) channels first, followed by the three-phase current ia. The channels ib(t), ic(t), and ic(t) are sequentially concatenated along the channel dimension to construct a three-dimensional time-frequency and channel tensor model with dimensions of "time scale × frequency scale × 6". The first dimension of the tensor is the time dimension, which carries the temporal correlation information of each sampling moment. The second dimension of the tensor is the frequency dimension, which carries the frequency resolution and sparse characterization information of the full-band harmonic components. The third dimension of the tensor is the observation channel dimension, which carries the channel transmission characteristic information of the 6 electrical quantity signals. This tensor completely preserves the full-dimensional characteristic information of the residual signal, which serves as standardized input data for subsequent harmonic source number estimation, hybrid matrix identification, and underdetermined separation processing.

[0028] The method for estimating the number of harmonic sources and identifying the mixing matrix based on tensor CP decomposition, such as... Figure 4 As shown, the implementation specifically includes: for the three-dimensional time-frequency and channel tensor with dimensions of "time scale × frequency scale × 6" that has been constructed, firstly, the kernel norm minimization is used to perform low-rank approximation processing on the tensor to suppress the time-frequency domain noise and the small aliasing components remaining from the sparse representation in the tensor, while completely preserving the main characteristic components corresponding to the effective harmonic sources in the tensor; then, based on the three-dimensional tensor after low-rank approximation, the Bayesian information criterion is used to traverse and calculate the criterion values ​​corresponding to different rank assumptions within the interval from 1 to the preset maximum number of harmonic sources, and the rank corresponding to the minimum value of the Bayesian information criterion is selected as the optimal rank of the tensor. This optimal rank is the actual number of harmonic sources N under the same measurement point. The value of N can be much larger than the number of observed channels 6, which is fully adapted to the actual working condition of having dozens of power electronic disturbance sources under the same measurement point; subsequently, a high-order orthogonal iteration is used to perform CANDECOMP / PARAFAC decomposition on the three-dimensional time-frequency and channel tensor that has completed low-rank approximation, decomposing the three-dimensional tensor into a linear superposition of N rank-1 tensors, and the iteration convergence threshold is set to The maximum number of iterations was set to 1000 to ensure the convergence and numerical stability of the decomposition results. Finally, based on the output of the CANDECOMP / PARAFAC decomposition, the factor matrix corresponding to the observation channel dimension was extracted. This factor matrix is ​​a 6×N mixing matrix. Each column of the mixing matrix corresponds to the transmission feature vector of a harmonic source on the 6 observation channels, and the column vectors are linearly independent, satisfying the necessary and sufficient condition for underdetermined blind source separation. This provides the core input parameters for subsequent sparse reconstruction and time-domain separation of single harmonic source components.

[0029] The method employs orthogonal matched pursuit to complete sparse reconstruction, restoring the full-band time-domain signals of each harmonic source, such as... Figure 4 As shown, the specific implementation includes: using the 6×N hybrid matrix identified in the previous steps as the dictionary matrix for sparse reconstruction; using the time-frequency sparse representation matrices corresponding to the 6 residual observation signals output in the previous steps as the reconstruction target; and for the observation data at each time-frequency point, performing sparse solution using orthogonal matching pursuit, with the iteration termination condition set as the L2 norm of the residual being less than 1. The iteration count is increased until the estimated number of harmonic sources N is reached, ensuring the convergence and sparsity constraints of the solution process. Through time-frequency point-by-time sparse solution, time-frequency domain sparse coefficient matrices corresponding to each of the N harmonic sources are obtained, which perfectly match the dimensions of the input time-frequency sparse representation matrix. The time-frequency domain sparse coefficient matrix of each harmonic source fully carries the time-frequency characteristic information of the full-band harmonic components corresponding to the disturbance source. For the time-frequency domain sparse coefficient matrix corresponding to each harmonic source, an inverse transformation process that perfectly matches the multi-scale variable window S-transform in the previous steps is performed. During the inverse transformation, the Gaussian window length parameter and time-frequency characteristic parameter corresponding to the original transformation are strictly followed. The frequency-scale mapping rule is used to restore the residual time-domain components of each harmonic source on the six observation channels. These residual time-domain components completely preserve the full-band harmonic information of the corresponding harmonic source, including its subharmonics, interharmonics, and integer higher harmonics. The fundamental wave components of the six observation signals obtained by fitting the recursive least squares algorithm in the previous steps are linearly weighted according to the proportion of the fundamental wave transmission characteristics of each harmonic source carried by the corresponding column vector in the mixing matrix. The allocated fundamental wave components are then superimposed on the residual time-domain components of the corresponding harmonic sources, finally obtaining complete time-domain signals of N harmonic sources, each independent of the others and including both fundamental wave and full-band harmonic components.

[0030] The accuracy of the verification separation result reconstruction, such as Figure 5 As shown, the specific implementation includes: first, acquiring the complete time-domain signals of the N harmonic sources separated in the previous steps, and then linearly superimposing them according to the identified mixing matrix to obtain the reconstructed observation signal; next, calculating the mean square error between the reconstructed observation signal and the original standardized synchronous observation signal output in the previous steps, and then comparing the calculated mean square error with a preset threshold. For comparison, when the mean square error is ≤ If the reconstruction accuracy of the harmonic separation result meets the requirements, the separation result is valid; if the mean square error is greater than 1, the result is considered valid. Then return to the previous steps to re-perform the multi-scale variable window S-transform and construct the three-dimensional tensor of the residual signal until the mean square error of the reconstructed observation signal and the original normalized synchronous observation signal satisfies ≤ The required precision.

[0031] The process of extracting harmonic features from each harmonic source and matching them with a typical power electronic disturbance source feature library for calibration includes the following steps: First, performing Fast Fourier Transform (FFT) spectral analysis on the complete time-domain signal of each harmonic source obtained in the previous steps to accurately extract the frequency, amplitude, and phase characteristic parameters of the fundamental, second harmonic, interharmonic, and integer harmonics of the harmonic source, forming a unique harmonic feature vector for that harmonic source; Second, constructing a typical power electronic disturbance source inherent harmonic emission feature library including photovoltaic inverters, charging piles, and energy storage converters. This feature library pre-stores the standard frequency range, amplitude threshold, and phase characteristic interval inherent characteristic parameters of the fundamental and full-band harmonics of each typical disturbance source, where the proportion of interharmonic amplitude corresponding to a specific frequency interval of the photovoltaic inverter is specified. The system identifies the amplitude characteristics of integer harmonics corresponding to specific orders in charging piles and the coordinated distribution patterns of second and higher harmonics in energy storage converters. Using the minimum Euclidean distance criterion, the harmonic feature vector of each harmonic source is compared with the inherent feature parameters of each typical disturbance source in the feature library. The disturbance source with the highest similarity and meeting the preset similarity threshold (set above 0.9) is selected as the matching object corresponding to that harmonic source. This completes the one-to-one calibration of each harmonic source with the corresponding disturbance source in the photovoltaic inverter, charging pile, and energy storage converter. Simultaneously, the disturbance source type, harmonic feature parameters, and matching similarity of each harmonic source are recorded to ensure the accuracy and traceability of the calibration results, providing data support for the accurate identification and control of harmonic sources in subsequent electricity metering.

[0032] The final output separation result, in practice, includes: integrating the complete time-domain signals of the N harmonic sources separated in the previous steps, the frequency, amplitude, and phase characteristic parameters of the fundamental wave and full-band harmonics (subharmonics, interharmonics, and integer higher harmonics) of each harmonic source extracted in the previous steps, and the one-to-one calibration results of each harmonic source with typical power electronic disturbance sources such as photovoltaic inverters, charging piles, and energy storage converters, including the disturbance source type and matching similarity corresponding to each harmonic source. At the same time, the validity identifier of this harmonic separation result and the corresponding mean square error value are recorded. All the above information is organized according to a standardized data format to form a complete harmonic separation and disturbance source calibration report, which is output to the storage module and data interaction interface of the power metering terminal, providing standardized and traceable final data support for accurate harmonic metering, disturbance source tracing and control in power metering.

[0033] Figure 6The comparison of the time-domain waveforms of the signals before and after separation with the reconstruction error provides intuitive visual support for verifying the reconstruction accuracy. The main coordinate system displays the time-domain waveforms of the original standardized observation signal and the reconstructed signal over 10 power frequency cycles; the two almost completely overlap, visually demonstrating the method's distortion-free preservation capability of the fundamental frequency and harmonic components across the entire frequency band. The secondary coordinate system displays the absolute reconstruction error curve, and the calculated mean square error of reconstruction (MSE) is 3.72 × [missing value]. Far below the preset 1× The accuracy threshold verified the effectiveness of the separation results. This figure also verifies the feasibility of the core technology chain of this application, proving that this method can overcome the underdetermined scenario limitations of traditional methods, achieve accurate separation and high-fidelity reconstruction of multiple harmonic sources under 6 observation channels, and solve the industry pain points of weak harmonic component submersion and component aliasing.

[0034] Figure 7 This chart shows a comparison of the full-band harmonic amplitude spectrum before and after separation. The horizontal axis covers the entire frequency band from 0 to 2500 Hz, and the vertical axis uses a logarithmic amplitude coordinate system based on the fundamental frequency, clearly displaying weak harmonic components as low as -80 dB. In the original signal amplitude spectrum, the strong fundamental frequency of 0 dB completely drowns out the subharmonics, interharmonics, and higher harmonics with amplitudes below -40 dB. After fundamental frequency stripping using the RLS method of this invention, the residual signal amplitude spectrum fully presents the harmonic characteristics of the entire frequency band, with no fundamental frequency residue or component aliasing. The separated single harmonic source amplitude spectrum accurately restores the inherent harmonic emission characteristics of charging piles and photovoltaic inverters, with a relative error of harmonic component extraction ≤1.8%, providing reliable support for the characteristic matching and calibration of disturbance sources.

[0035] This application also provides an embodiment of an electronic device. The electronic device is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors or processing units, memory, and buses connecting different components (including memory and processing units).

[0036] A bus refers to one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MCA) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0037] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0038] The memory may include computer-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. Electronic devices may further include other removable / non-removable, volatile / non-volatile computer device storage media. By way of example only, the storage system may be used to read and write non-removable, non-volatile magnetic media.

[0039] The electronic device can also communicate with one or more external devices (e.g., keyboard, pointing device, camera, etc.), may include a display, and may communicate with one or more devices that enable a user to interact with the electronic device, and / or with any device that enables the electronic device to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via an input / output (I / O) interface. Furthermore, the electronic device can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. The processor executes various functional applications and data processing by running programs stored in memory, such as implementing the harmonic multidimensional component separation method for power metering provided in the above embodiments of the present invention.

[0040] This application also discloses a harmonic multi-dimensional component separation system for electricity metering, including: The signal preprocessing module is used to synchronously acquire three-phase voltage and current observation signals at a fixed frequency, complete power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization, and output standardized synchronous observation signals. The fundamental wave stripping module is used to fit and strip the strong fundamental wave component using the recursive least squares method to obtain a residual signal containing harmonics across the entire frequency band. The time-frequency tensor construction module is used to perform multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components and construct three-dimensional time-frequency and channel tensors. The harmonic source separation module is used to estimate the number of harmonic sources and identify the mixing matrix based on tensor CP decomposition. It then uses orthogonal matching pursuit to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. The result verification and calibration module is used to verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

[0041] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for separating multidimensional harmonic components in electricity metering, characterized in that, Includes the following steps: Three-phase voltage and current observation signals are synchronously acquired at a fixed frequency. Power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization are completed, and standardized synchronous observation signals are output. The recursive least squares method was used to fit and remove the strong fundamental wave component to obtain the residual signal containing harmonics across the entire frequency band. Perform a multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components, and construct a three-dimensional time-frequency and channel tensor; Based on tensor CP decomposition, the number of harmonic sources is estimated and the mixing matrix is ​​identified. Orthogonal matching pursuit is used to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. Verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

2. The method for separating harmonic components in power metering according to claim 1, characterized in that, The process of completing power frequency synchronization alignment, outlier removal, DC bias removal, and amplitude normalization to output a standardized synchronous observation signal includes: using a digital phase-locked loop to perform power frequency period synchronization alignment based on zero-crossing detection on the acquired multi-channel observation signals; using a corresponding criterion to remove pulse-type outliers from the signals; using linear interpolation of adjacent effective sampling points to complete the corresponding missing data; using a moving average filter to remove the DC bias component from the signals; performing amplitude normalization on the signals that have undergone the aforementioned processing; and finally outputting a standardized multi-channel synchronous observation signal.

3. The method for separating harmonic multidimensional components in power metering according to claim 1, characterized in that, The process of fitting and stripping strong fundamental components using the recursive least squares method to obtain a residual signal containing full-band harmonics includes: using the synchronous phase-locked loop power frequency phase corresponding to the standardized synchronous observation signal as a reference, constructing a time-domain fitting model containing the positive, negative, and zero-sequence components of the fundamental wave; using the recursive least squares algorithm to perform real-time fitting of the fundamental wave components of multiple standardized synchronous observation signals respectively; subtracting the corresponding fitted fundamental wave component from each original standardized synchronous observation signal to obtain the residual observation signal after fundamental wave stripping; and the residual signal completely retaining the full-band harmonic components in the original signal.

4. The method for separating harmonic multidimensional components in power metering according to claim 1, characterized in that, The step of performing a multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components includes: performing time-frequency domain transformation on the multi-path residual observation signals output from the previous step using a multi-scale variable window S-transform; adaptively matching Gaussian windows of corresponding window lengths for transformation processing on subharmonics, interharmonics, and integer higher harmonics components in different frequency bands; and achieving sparse representation of harmonic components in the time-frequency domain across the entire frequency band through adaptive adjustment of the Gaussian window width. This enables harmonic components of different frequencies and modulation types to form a unique single-peak sparse representation on the corresponding time-frequency scale, eliminating component aliasing caused by cross-dimensional harmonic modulation.

5. The method for separating harmonic multidimensional components in power metering according to claim 4, characterized in that, The construction of the three-dimensional time-frequency and channel tensor includes: performing dimensional consistency verification and alignment on the time-frequency sparse representation matrix output by the residual observation signal after multi-scale variable window S-transformation to ensure that the time scale dimension and frequency scale dimension of the matrix corresponding to each signal are completely matched; then, taking the observation channel as the third dimension, the dimensionally aligned time-frequency sparse representation matrix is ​​sequentially spliced ​​along the channel dimension to construct a three-dimensional time-frequency and channel tensor model, which fully preserves the full-dimensional feature information of the residual signal and serves as standardized input data for subsequent harmonic separation processing.

6. The method for separating harmonic multidimensional components in power metering according to claim 1, characterized in that, The method for estimating the number of harmonic sources and identifying the mixing matrix based on tensor CP decomposition includes: for the constructed three-dimensional time-frequency and channel tensors, firstly, the tensor is approximated with low rank by minimizing the kernel norm to suppress the aliasing components of time-frequency domain noise and sparse representation residues, while fully preserving the main characteristic components of effective harmonic sources; then, based on the three-dimensional tensor after low-rank approximation, the optimal rank of the tensor is determined by the Bayesian information criterion to obtain the actual number of harmonic sources at the measurement point; subsequently, high-order orthogonal iteration is used to perform CP decomposition on the tensor, decomposing it into a linear superposition of rank-1 tensors matching the number of harmonic sources; finally, the factor matrix of the observation channel dimension is extracted based on the decomposition results as the mixing matrix, providing the core input for subsequent sparse reconstruction and time-domain separation of single harmonic source components.

7. The method for separating harmonic multidimensional components in power metering according to claim 6, characterized in that, The process of using orthogonal matched pursuit to complete sparse reconstruction and restore the full-band time-domain signal of each harmonic source includes: using the hybrid matrix identified in the previous step as the dictionary matrix for sparse reconstruction; using the time-frequency sparse characterization matrix corresponding to the residual observation signal output in the previous step as the reconstruction target; performing sparse solution point by point using orthogonal matched pursuit to obtain the time-frequency domain sparse coefficient matrix corresponding to each harmonic source; performing an inverse transform on the coefficient matrix that matches the previous multi-scale variable window S-transform to restore the residual time-domain components of each harmonic source; and then allocating the fundamental wave full component according to the proportion of the fundamental wave transmission characteristics of each harmonic source carried by the hybrid matrix and superimposing them to finally obtain the complete time-domain signal of each harmonic source containing the fundamental wave and the full-band harmonic components.

8. The method for separating harmonic multidimensional components in power metering according to claim 1, characterized in that, The verification of the reconstruction accuracy of the separation result includes: acquiring the complete time-domain signals of each harmonic source obtained from the previous steps, linearly superimposing the identified hybrid matrix to obtain the reconstructed observation signal, calculating the error between the reconstructed observation signal and the original standardized synchronous observation signal output from the previous steps, comparing the error with a preset threshold, and determining that the reconstruction accuracy of the separation result meets the threshold requirement and the result is valid if the error does not meet the threshold requirement. Otherwise, returning to the corresponding previous steps for reprocessing is carried out until the error meets the preset accuracy requirement.

9. The method for separating harmonic multidimensional components in power metering according to claim 8, characterized in that, The step of extracting harmonic features of each harmonic source and matching them with a feature library of typical power electronic disturbance sources to complete calibration includes: performing spectral analysis on the complete time-domain signals of each separated harmonic source, extracting the feature parameters of the fundamental wave and the full-band harmonics, generating the harmonic feature vectors of the corresponding harmonic sources, constructing a feature library of inherent harmonic emission of typical power electronic disturbance sources, pre-storing the inherent feature parameters of the full-band harmonics of various disturbance sources, using a similarity calculation criterion to match the harmonic feature vectors of each harmonic source with the feature library parameters, completing the one-to-one calibration of the harmonic sources and their corresponding disturbance sources, and storing the relevant calibration data.

10. A system utilizing the harmonic multidimensional component separation method for power metering according to claim 1, characterized in that, include: The signal preprocessing module is used to synchronously acquire three-phase voltage and current observation signals at a fixed frequency, complete power frequency synchronization alignment, outlier removal, DC bias removal and amplitude normalization, and output standardized synchronous observation signals. The fundamental wave stripping module is used to fit and strip the strong fundamental wave component using the recursive least squares method to obtain a residual signal containing harmonics across the entire frequency band. The time-frequency tensor construction module is used to perform multi-scale variable window S-transform on the residual signal to achieve decoupling and sparse representation of cross-modulation components and construct three-dimensional time-frequency and channel tensors. The harmonic source separation module is used to estimate the number of harmonic sources and identify the mixing matrix based on tensor CP decomposition. It then uses orthogonal matching pursuit to complete sparse reconstruction and restore the full-band time domain signal of each harmonic source. The result verification and calibration module is used to verify the reconstruction accuracy of the separation results, extract the harmonic features of each harmonic source, match them with the feature library of typical power electronic disturbance sources to complete the calibration, and output the final separation results.

Images