CN114781196A - Harmonic detection method based on sparse acquisition model - Google Patents
Harmonic detection method based on sparse acquisition model Download PDFInfo
- Publication number
- CN114781196A CN114781196A CN202210698874.3A CN202210698874A CN114781196A CN 114781196 A CN114781196 A CN 114781196A CN 202210698874 A CN202210698874 A CN 202210698874A CN 114781196 A CN114781196 A CN 114781196A
- Authority
- CN
- China
- Prior art keywords
- harmonic
- phasor
- frequency
- matrix
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
- G01R23/16—Spectrum analysis; Fourier analysis
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a harmonic detection method based on a sparse acquisition model, which comprises the following steps of: establishing a broadband dynamic signal model, and calculating an attenuation direct current component; estimating harmonic phasor in the broadband dynamic signal: establishing a harmonic phasor estimation model, reconstructing the harmonic phasor estimation model, and solving the harmonic estimation model; simulation analysis: and establishing a test scene, and sequentially establishing a basic performance test scene, a frequency slope test scene, a step change test scene and an interference test scene. The method has higher precision and speed for estimating the broadband dynamic phasor, keeps certain superiority in a comparison algorithm, and simultaneously verifies the dynamic performance and the anti-interference capability of the method.
Description
Technical Field
The invention relates to the field of electronic information, in particular to a harmonic detection method based on a sparse acquisition model.
Background
With the rapid development of smart grids and new energy power generation, a large number of nonlinear devices are applied to power systems, resulting in the increase of inter-harmonic and DDC components in the grids. On one hand, the harmonic waves and the inter-harmonic waves have frequency spectrum interference on the concerned frequency, and the accurate metering of the electric energy is influenced. On the other hand, aiming at the attenuated direct current component in the electric signal when the power system fails, the traditional algorithm cannot filter the attenuated direct current component, so that the error of the calculation result is large. Therefore, an effective phasor estimation algorithm needs to be found to accurately measure the broadband dynamic phasor.
At present, phasor measurement methods are mainly classified into two types: DFT algorithms (discrete fourier transform) and non-DFT algorithms. The DFT algorithm is to sample the frequency spectrum of a finite time domain discrete signal at equal intervals and discretize a frequency domain function. When the traditional DFT algorithm is adopted to measure phasor, the problems of information redundancy, mutual harmonic interference, information leakage and the like can occur when the actual frequency deviates, and a larger measurement error is caused. Some studies propose to introduce a zero-crossing detection method, a wavelet transform method, an instantaneous value method and the like to optimize the DFT algorithm, but the methods still have large errors at non-nominal frequencies. The related document describes the use of improved second harmonic filtering techniques for monophasic phasor measurements at non-nominal frequencies. The method can integrate uniform sampling and fixed window length phasor measurement without changing the structure of the traditional phasor processing method, and has good performance. However, the phasor estimation algorithm based on DFT under the static model has the problem of poor dynamic performance.
For the bad point that the cracked DTF model is only limited to static signal analysis, some research proposals develop dynamic phasor estimation to different mathematical frameworks to obtain a new generation phasor measurement method based on the model. The method applies a specific mathematical model to define the amplitude and phase changes of the signal in a stable sampling time interval, and gets rid of the limitation of the period in Fourier analysis.
Currently, dynamic harmonic analysis methods, such as taylor-fourier transform (TFT), can achieve phasor estimation in the case of power oscillations, but are susceptible to interference from inter-and higher harmonics. The invention widens the frequency range of the TFT and overcomes the limitation of DFT periodicity. However, the method has many and complex system parameters, and has a problem of large error after multiple fitting, and the measurement error increases with the increase of the order. In order to greatly improve the estimation accuracy, the related literature proposes a discrete fourier transform interpolation method with high resolution. The invention has the capability of recovering the stage information and achieves the noise suppression by blocking the iteration times. However, the DFT sequence in this method can only be resampled at the nominal rate, which undoubtedly adds to the uncertainty of the effect.
Based on the methods, matrix transformation coefficients are obtained through high-order matrix inversion operation. Despite the many effective error reduction algorithms, the effects of higher order inter-harmonics and noise superposition cannot be addressed. Thus, consider incorporating a machine learning algorithm to reduce phasor solution computational complexity. The related literature effectively tracks harmonics for modeling estimation by identifying the most relevant components in the signal, thereby limiting the effects of inter-harmonics. However, this method is only suitable for low-band harmonic vectors, and the phasor estimation parameters cannot be directly obtained because the high-order harmonic signals have time-varying property and the solution of the high-order pseudo-inverse matrix has high complexity. And further considering the field of dynamic harmonic analysis, related documents perform iterative estimation on harmonic parameters by using an STWLS algorithm, and compensate the influence of the harmonic parameters on the estimation. The method can improve the frequency measurement precision, allocate more space for dynamic information, and still be difficult to resist the interference of constraining the inter-harmonic. The related literature proposes O-spline dynamic harmonic analysis, and the method reduces the calculation complexity of the DTTFT in a closed form and provides an optimal data compression algorithm for oscillation. However, under non-ideal conditions, inter-harmonic interference becomes more and more severe as the order increases due to the presence of large noise in the spatial step reconstruction process.
In order to solve the problem of solving the synchronous phasor estimation, a published related document proposes complex wavelet transform based on a Fast Fourier Transform (FFT) algorithm, which is used for segmenting an original waveform and analyzing signal reconstruction. However, in continuous multi-scale analysis, this may result in severe phase distortion and loss of timing information for further signal processing. There is also a published relevant document to obtain frequency, amplitude and phase estimates in harmonics by improving the Taylor weighted least squares algorithm. The method is not suitable for high-order harmonic measurement because the method is only suitable for measuring third-order continuous waves and cannot be used for high-precision harmonic analysis.
Disclosure of Invention
Based on the technical problem, the invention provides a harmonic detection method based on a sparse acquisition model, which comprises the steps of firstly carrying out parameterized modeling on an attenuation direct current component by utilizing a sinc interpolation function, obtaining an accurate DDC component through a least square method, then adopting an improved Taylor-Fourier model for broadband harmonic component estimation, combining a machine learning algorithm, accurately recovering a specific signal by using fewer data points, and further converting a harmonic phasor reconstruction problem into a sparse acquisition model solving problem. Under the regularization frame, the invention expresses the phasor reconstruction problem into a form of a minimization function, and the underdetermined inverse problem is effectively solved by utilizing alternate minimization operation under an SBI system, so that the harmonic phasor estimation value is efficiently and accurately obtained. To prove the rationality of the model, the error range of the estimated values is measured using a cross-entropy objective function. Simulation results show that the method has higher precision and speed for estimating the broadband dynamic phasor, certain superiority is kept in a comparison algorithm, and the dynamic performance and the anti-interference capability of the method are verified.
The technical scheme of the invention is as follows: a harmonic detection method based on a sparse acquisition model comprises the following steps:
a. estimating the attenuation direct current component in the broadband dynamic signal: establishing a broadband dynamic signal model, and calculating an attenuated direct current component;
b. estimating harmonic phasor in the broadband dynamic signal: establishing a harmonic phasor estimation model, reconstructing the harmonic phasor estimation model, and solving the harmonic estimation model;
c. a simulation analysis step: and establishing a test scene, and sequentially establishing a basic performance test scene, a frequency slope test scene, a step change test scene and an interference test scene.
In step a, the step of establishing a broadband dynamic signal model is as follows:
for power system waveforms containing a DC component and a fundamental componentThe expression is shown as (1):
wherein, the first and the second end of the pipe are connected with each other,is a signal containing fundamental, harmonic and inter-harmonic components,is a time-varying amplitude of the signal,is a time-varying phase angle, the reference frequency isThe term "refers to the highest order harmonic component,is to attenuate the signal of the direct current component,is the amplitude of the attenuated dc component,is a time constant;
signals containing fundamental, harmonic and inter-harmonic componentsThe dynamic phasor model at time t is:
substituting equation (2) into equation (1) to obtain broadband dynamic signalThe dynamic expression of (2):
In step a, the step of "calculating the attenuated dc component" is as follows:
attenuating DC componentRepresented by the sum of the amplitude and phase time-varying cosine components:
wherein, the first and the second end of the pipe are connected with each other,,,respectively representing the frequency, amplitude and phase of the nth cosine component, T is the length of an observation interval, and Y is set as3;
Based on the frequency domain sampling theorem, the time domain signal parameterization modeling is carried out on the attenuation direct current component by utilizing a sinc interpolation function, and the dynamic phasor of the DDC component is recorded asSpecifically, it is represented as:
wherein, the first and the second end of the pipe are connected with each other,is the sampling frequency, represents the maximum model order, will be set to 2,is the nth cosine component in the formula (2) at the timeThe sample phasor of (c);
for attenuating DC componentSampling is carried out, and the discrete form of the fitted DDC component in the formula (1) is represented by (6);
wherein the content of the first and second substances,representing phasorsIs a column vector of the cosine signal,as conjugate operators, matricesIs an element ofAt a sample point in the sampling window(s),representation matrix,The definition of (2) is shown as (7):
determining phasor estimation values by means of least squaresThe calculation process is shown in (7) in the matrixIn (1),for the matrix elements:
wherein H is a conjugate transposed symbol;
by constructing each cosine component using equation (8)Followed byReconstruction matrixThe amplitude and the phase of each cosine signal are obtained through (5), and then DDC components are estimated。
In step b, the step of establishing a harmonic phasor estimation model is as follows:
the dynamic phasors are defined using a taylor expansion model of order:
wherein, the first and the second end of the pipe are connected with each other,is t =0K is the taylor expansion order, and T is the observation interval length;
the sinusoidal signals containing harmonics and inter-harmonics take the form of discrete sequencesThe sequence length is N, andand expanding the dynamic phasor method to an actual sequence model through multi-frequency phasor analysis, wherein the Taylor Fourier coefficient expression of each phasor component of the discrete signal is as follows:
where H is the sampling interval, observation interval length,Is the average harmonic and inter-harmonic phasors,is at phasor frequencyThe derivative of the order k of (c) is,is the fundamental frequencyInteger multiples of (d) represent harmonic frequencies, and non-integer multiples represent inter-harmonic frequencies;
in a size ofThe sampling rate of (2) normalizes the frequency of the harmonic signal by a sampling length ofObtaining the frequency of the harmonic componentThe normalized frequency of the Taylor Fourier basis vector is,The frequency resolution corresponding to the N coefficient sets isAnd the taylor opening order of each i is K, then (10) is written in matrix form:
wherein the content of the first and second substances,is a sample column phasor of sizeIs a Taylor Fourier-based phasor, e represents noise or measurement error, and x is a length ofThe column phasors of (1), describe the current harmonicsWhen the utility model is used, the water is discharged,a set of (a);
and (3) phasor solving, namely converting the phasor solving into solving of a pseudo-inverse matrix by using a least square method, and calculating the pseudo-inverse matrix to obtain a corresponding solution when the Euclidean norm is minimum, wherein the solution is the optimal solution under the following constraint conditions:
wherein, the first and the second end of the pipe are connected with each other,expressing Euclidean norm, and further obtaining a dynamic phasor coefficient estimation matrix by introducing a Lagrange operator and performing derivation:
In step b, the step of reconstructing the harmonic phasor estimation model is as follows:
setting a correction factorEach group of frequency resolution reachesThe length of the sampling sequence isThe sampling length H is 5 fundamental wave periods, and the harmonic component normalized frequency is,And obtaining an improved Taylor Fourier coefficient expression:
wherein the content of the first and second substances,,is expressed as a size ofThe matrix of the curvelet transform of (c),is thatThe k-th derivative of (a);
according to the expression (9), an approximate second-order dynamic phasor estimation value and a harmonic frequency are obtainedAnd frequency rate of change:
When correcting the factorWhen the sensing matrix A and the sparse matrix are in useHighly uncorrelated, (14) is converted into matrix form, namely:
wherein, the first and the second end of the pipe are connected with each other,is a measurement matrix, x represents a length ofFor each data block ofIn the present sequence, the sequence of the method,is the superposition of a small number of phasors extracted after the original sample is segmented,is dimension ofEach column of the matrix is a base phasor of a discrete fourier transform, e represents a noise signal phasor;
after the sparse sampling measurement model is constructed, matrix blocking is carried out on x by using an Euclidean search algorithm, m data points with optimal matching are found in a search range, and all the data points are combined into a matrixI is a coordinate of the coefficient matrix, curvelet transformation is carried out on the matrix, an insignificant curvelet coefficient is removed by adopting a curvelet threshold criterion, an effective denoising and compression algorithm is required for the extracted coefficient, curvelet transformation is carried out, an insignificant curvelet coefficient is removed by adopting a curvelet threshold criterion, and an effective denoising and compression algorithm is required for the extracted coefficient;
sparse matrixOnly contains a non-zero significant term set, the phasor of the transformation coefficient column obtained after the significant term set is arranged according to the dictionary order isOf the transform coefficientsCentered around the zero region and distributed in the form of a fine peak, the signal repeatability is characterized using the distribution characteristics of the transform coefficients, noted as:
each set of harmonic signal estimates may be inverted according to the transform coefficients:
wherein, the first and the second end of the pipe are connected with each other,is thatThe inverse operator of (2).
In step b, the step of solving the harmonic estimation model is as follows:
solving a regularized linear least square cost function, and reconstructing a Taylor coefficient matrix x, namely:
wherein, the first and the second end of the pipe are connected with each other,the norm of L2 is shown,a regularization term is represented as a function of,is a coefficient of the regularization that,、representing local and non-local sparse terms respectively,is a regularization parameter that balances the two sparse terms;
(22) substituting (21) to obtain:
introducing two auxiliary phasors p and q, finally the following scheme is obtained:
wherein, the first and the second end of the pipe are connected with each other,andthe parameters are fixed value parameters, the function is to improve the stability of the algorithm value, and p and q are SBI algorithm auxiliary iteration phasors;
(24) solving the minimization of a strict convex quadratic function, setting the gradient of an objective function as 0, and obtaining a corresponding closed solution:
the minimization problem of the function is solved by adopting the steepest descent method with the optimal step length, which is expressed as follows:
wherein, the first and the second end of the pipe are connected with each other,is the gradient of the objective function and,is the optimum step size for the particular application,the estimated value isUsing the least mean square error theorem of (25)The problem becomes:
similarly, the question of (26) is written as:
Obtaining a set of closed solutions, which areObserved values seen as noiseOf a certain type, useRepresenting the error phasor, the error of each element is respectively() Then assume e is independently distributed over Mean 0 and varianceFrom law of large numbers, for arbitraryAll have:
so as to obtain:
the transformed error phasor is, Representing the error phasor for each element, m being the number of data points for the best match, and the transformation does not change the variance of each group according to the orthogonal nature of the matrix, from which it follows that for each groupAre all independently distributed in(zero mean and variance of) Most of them are theoretic and arbitraryAll have:
the method comprises the following steps:
to obtain:
combining the above and the (26) problem to obtain:
wherein the content of the first and second substances,refers to the modulus of the corresponding phasor;
under the vector contraction form, the estimated value of x is expressed as:
after the estimation value of the Taylor coefficient matrix x is obtained, the method is carried outHarmonic sampling equation to harmonic signalsIn the framework of logistic regression, cross entropy was introduced [27 ]]The objective function measures the difference in probability distribution between the estimated and theoretical values x:
assuming that the error is a binary distribution,the predicted probability distribution is considered to be closely related to the actual probability distribution, proving that the assumption is consistent with the expected model.
In step c, the step of establishing a basic performance test scene is as follows:
the basic performance of the proposed algorithm is tested, and a broadband dynamic signal model containing DDC components shown in formula (43) is constructed:
wherein the content of the first and second substances,for the fundamental frequency, here set at 50Hz,、respectively represent the fundamental wave and each harmonic phase angle, and are inThe value of the harmonic times h of the low frequency band is 2-13, the harmonic times h of the high frequency band are 77, 79, 80 and 83 respectively, the sampling frequency is set to be 10kHz, the amplitude of the DDC component is 0.3, 0.4, 0.5 and … 1, and the time constant of the DDC component is arbitraryStarting from 0.01s, changing from 0.01s to 0.1s in steps;
applying BMP algorithm, the initial value of the iterative phasor isRegularization coefficient ofFixed value parameter,, When the zeta value is in the range of [0.05,0.3 ]]Setting the number of the best matched data points in the search window as m;
FFT, Prony, TWLS, SIFE were chosen as comparison algorithms.
In step c, the step of establishing a frequency ramp test scenario is as follows:
the BMP was analyzed for performance under a frequency ramp, providing the following signals:
wherein the content of the first and second substances,is the fundamental frequency and takes the value of 50Hz,is the slope of the fundamental frequency, the value is 1Hz/s,anda fundamental phase and a harmonic phase, respectively, the phases being set toRandom numbers uniformly distributed within the range.
In step c, the step of establishing the step change test scene comprises the following steps:
at the start of the test, the amplitude of each component was set to 115% of the initial amplitude, and the phase was changed toThe broadband dynamic signal provided is as follows:
and setting the sampling rate to be 5kHz and the length of the sampling period to be 5 periods, and evaluating the performance of each algorithm under the condition of step change by using the speed of response time in the standard.
In step c, the step of establishing an interference test scene is as follows:
gaussian white noise with the signal-to-noise ratio of 55dB is introduced into the signal, and specific broadband dynamic signals are as follows:
wherein the content of the first and second substances,is the inter-harmonic frequency, and the values thereof are 9652.5Hz, 9751.5Hz, 9850.5Hz, 9949.5Hz, 10048.5Hz and 10147.5Hz in sequence,andis the phase of fundamental wave and harmonic wave, and takes the value ofThe random number in the range is set to have a sampling frequency of 10kHz and a sampling window length of 5 power frequency periods.
The invention has the beneficial effects that:
a. the method is divided into two parts, wherein the first part obtains accurate DDC components based on a least square method, the second part establishes a Taylor-Fourier model of broadband dynamic phasors, the regularization optimization problem of a sparse acquisition model is solved by utilizing a harmonic vector estimation method, and finally, an estimated value of harmonic phasor measurement is obtained by utilizing a segmented SBI iteration frame, so that the reconstruction of an original signal is realized, the phasor measurement and estimation precision is obviously improved through a simulation test and a performance test, and a reliable theoretical basis can be provided for PMU measurement;
b. the invention firstly uses a sinc interpolation function to carry out parametric modeling on an attenuation direct current component, obtains an accurate DDC component by a least square method, adopts an improved Taylor-Fourier model aiming at broadband harmonic component estimation, combines a machine learning algorithm, accurately recovers a specific signal by fewer data points, further converts a harmonic phasor reconstruction problem into a sparse acquisition model solving problem, under the condition of a regularization frame, the invention expresses the phase reconstruction problem into a form of a minimization function, and effectively solves the underdetermined inverse problem by using alternate minimization operation under an SBI system, thereby efficiently and accurately obtaining a harmonic phasor estimation value, in order to prove the rationality of the model, an error range of the estimation value is measured by using a cross entropy objective function, and a simulation result shows that the estimation of the broadband dynamic phasor has higher precision and speed, certain superiority is kept in the comparison algorithm, and the dynamic performance and the anti-interference capability of the algorithm are verified;
c. the invention provides a new broadband dynamic phasor measurement algorithm, which solves the problem of effective processing of broadband dynamic phasors containing DDC components; based on the idea of regularizing the sparse capture matrix, obtaining a phasor estimation result by solving a sparse regularization problem; simulation and experiment results show that key information of the broadband dynamic phasor is identified, and the calculation complexity is obviously reduced; this represents the ability to obtain more accurate results in a short time, thereby effectively detecting the transient characteristics of a broadband signal; meanwhile, under static and dynamic conditions such as noise interference, inter-harmonic interference, frequency slope and the like, the measurement result can meet the test requirement of the M-level PMU.
Drawings
FIG. 1 is a diagram of the reconstruction effect and the operating time range as a function of parameters;
FIGS. 2[ a ], 2[ b ], 2[ c ] are graphs comparing maximum errors of TVE, FE, RFE under the condition of frequency ramp, FIG. 2[ a ] is a graph comparing maximum errors of TVE under the condition of frequency ramp, FIG. 2[ b ] is a graph comparing maximum errors of FE under the condition of frequency ramp, and FIG. 2[ c ] is a graph comparing maximum errors of RFE under the condition of frequency ramp;
FIG. 3 is a graph of the ratio of run times for the methods at step down;
fig. 4 a, 4 b, and 4 c are maximum error maps of TVE, FE, and RFE under inter-harmonic interference, fig. 4 a is a maximum error map of TVE and RFE under inter-harmonic interference, fig. 4 b is a maximum error map of FE under inter-harmonic interference, and fig. 4 c is a maximum error map of RFE under inter-harmonic interference.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
Example (b):
1. estimation of attenuated DC components in broadband dynamic signals
1.1 broadband dynamic signal model
For power system waveforms containing a DC component and a fundamental componentThe expression is shown as (1):
whereinIs a signal containing fundamental, harmonic, and inter-harmonic components.Is a time-varying amplitude of the signal,is a time-varying phase angle, the reference frequency is,Refers to the highest order harmonic component.Is to attenuate the signal of the direct current component,is to attenuate the magnitude of the dc component,is a time constant.
Signals containing fundamental, harmonic and inter-harmonic componentsThe dynamic phasor model at time t is:
substituting equation (2) into equation (1) to obtain broadband dynamic signalThe dynamic expression of (2):
1.2, attenuated DC component estimation
Attenuating DC componentCan be represented by the sum of the amplitude and phase time-varying cosine components:
wherein, the first and the second end of the pipe are connected with each other,,,respectively representing the frequency, the amplitude and the phase of the nth cosine component. And T is the observation interval length. To meet the accuracy requirement, the present embodiment sets Y to 3.
Based on the frequency domain sampling theorem, the time domain signal parameterization modeling is carried out on the attenuation direct current component by utilizing a sinc interpolation function, wherein the dynamic phasor of the DDC component is recorded asSpecifically, it is represented as:
wherein the content of the first and second substances,is the frequency of the sampling of the samples,represents the maximum model order, and the present embodiment willSet to 2.Is the nth cosine component of the formula (2) at the timeThe sample phasor above.
For attenuating DC componentSampling is performed, and the discrete form of the fitted DDC component in equation (2) can be represented by (5).
WhereinRepresenting phasorsIs the column vector of the cosine signal.Is a conjugate operator. Matrix ofIs an element ofA sample point in a sampling window.Representation matrix,The definition of (2) is shown as (7):
determining phasor estimation values by means of least squares. The calculation process is shown as (9). In a matrixIn (1),is the matrix element.
Where H is the conjugate transposed symbol.
Each cosine component being constructed using equation (8). Followed byReconstruction matrix. The amplitude and the phase of each cosine signal are obtained through (5), and then DDC components are estimated。
2. Harmonic phasor estimation in broadband dynamic signals
2.1 establishment of harmonic phasor estimation model
It is contemplated that the taylor fourier method can describe the amplitude and phase transformation over time within the observation interval. Thus, the dynamic phasors are defined using a taylor expansion model of order:
wherein the content of the first and second substances,is t =0K is the taylor expansion order, and T is the observation interval length.
In practice, the sinusoidal signals containing harmonics and inter-harmonics are generally in the form of discrete sequencesThe sequence length is N, and. Through multi-frequency phasor analysis, a dynamic phasor method is expanded to an actual sequence model, Taylor Fourier coefficient expressions of each phasor component of a discrete signal are as follows:
where H is the sampling interval, observation interval length。Is the average harmonic and inter-harmonic phasors,is at phasor frequencyThe k-th order derivative of (c).Is the fundamental frequencyInteger multiples of (d) represent harmonic frequencies, and non-integer multiples represent inter-harmonic frequencies.
In the size ofThe sampling rate of (2) normalizes the frequency of the harmonic signal by a sampling length ofObtaining the frequency of the harmonic componentThe normalized frequency of the Taylor Fourier basis vector is,. The frequency resolution corresponding to the N coefficient sets is. Considering that the taylor opening order of each i is K, (11) can be written in the form of a matrix:
wherein the content of the first and second substances,is the sample column phasor. A size ofIs a taylor fourier-based phasor. e represents noise or measurement error. x is a length ofThe column phasors of (1), describe the current harmonicsWhen the utility model is used, the water is discharged,a collection of (a).
Usually, the phasor solution can be converted into a solution of a pseudo-inverse matrix using a least squares method. The solution corresponding to the minimum Euclidean norm is obtained through the calculation of the pseudo-inverse matrix, and the solution is the optimal solution meeting the following constraint conditions:
wherein the content of the first and second substances,representing the euclidean norm. Further, by introducing Lagrange operator and performing derivation, a dynamic phasor coefficient estimation matrix can be obtained:
However, as the maximum harmonic order increases, the matrix dimension increases rapidly, and the amount of computation required to solve the multi-element linear equation corresponding to equation (14) increases significantly. For high frequency harmonics, the phasor method based on the least squares method is the least computationally complex. Therefore, the phasor method has a problem of low efficiency in solving the multidimensional matrix. In addition, since the observation time and the frequency resolution are contradictory, the frequency analysis is not accurate enough, and the time resolution is low.
2.2 reconstruction of harmonic phasor estimation model
To obtain more accurate frequency domain results, the present embodiment introduces a correction factor in the harmonic phasor analysis. Each group of frequency resolution reachesThe length of the sampling sequence isThe sampling length H is 5 fundamental periods. The harmonic component normalized frequency can be expressed as,。
Improved taylor fourier coefficient expressions are obtained under more precise correction conditions:
here, the,Is expressed as a size ofThe matrix of the curvelet transform of (c),is thatThe k-th derivative of (c).
According to the expression (9), an approximate second-order dynamic phasor estimated value and a harmonic frequency can be obtainedAnd frequency rate of change:
Since the number of harmonic unknowns is much larger than the observed values, it is assumed that x is compressible in the curvelet transform domain. When correcting the factorWhen the sensing matrix A and the sparse matrix are in useHighly uncorrelated, (14) is converted into matrix form, namely:
whereinIs a measurement matrix. x represents a length ofThe corresponding sample sequence of each data block of (a),is the superposition of a small number of phasors extracted after the original sample is segmented.Is dimension ofEach column of the matrix is a fundamental phasor of a discrete fourier transform. e represents the noise signal phasor.
After the sparse sampling measurement model is constructed by the method, the matrix partitioning is performed on x by using the Euclidean search algorithm based on the sparsity of harmonic frequency domain distribution. Finding m best-matching data points within the search range and combining all data points into a matrixAnd i is the coordinate of the coefficient matrix. And performing curvelet transformation on the matrix, removing insignificant curvelet coefficients by adopting a curvelet threshold criterion, wherein the extracted coefficients need effective denoising and compression algorithms. And (4) performing curvelet transformation, and removing insignificant curvelet coefficients by adopting a curvelet threshold criterion, wherein the extracted coefficients need an effective denoising and compression algorithm.
To ensure locally smooth features for harmonic detection, sparse matricesOnly a small set of non-zero valid entries is included. The phasor of the transformation coefficient column obtained after the effective item set is arranged according to the dictionary order is. Due to the transformation of coefficientsThe distribution of the transform coefficients is concentrated near the zero region and distributed in the form of a fine peak, so that the repeatability of the signal is characterized by using the distribution characteristics of the transform coefficients, and is recorded as follows:
Each set of harmonic signal estimates may be inverted according to the transform coefficients:
2.3 solution of harmonic estimation model
In order to obtain an estimate with reasonable accuracy and noise robustness, after a harmonic phasor recovery model is established based on the above method, the following taylor coefficient matrix x is reconstructed by solving an optimization problem of a regularized linear least squares cost function, that is:
wherein, the first and the second end of the pipe are connected with each other,representing the L2 norm.A regularization term is represented as a function of,is a system of regularizationAnd (4) counting.、Representing local and non-local sparse terms, respectively.Is a regularization parameter that balances the two sparse terms.
(22) Substituting (21) to obtain:
To solve the above minimization problem, the present embodiment employs an alternative SBI algorithm. Introducing two auxiliary phasors p and q, the following scheme is finally obtained:
wherein the content of the first and second substances,andthe parameter is a fixed value parameter and has the function of improving the stability of the algorithm value. p, q are the SBI algorithm assisted iterative phasors.
Equation (24) is a problem of solving the minimization of a strictly convex quadratic function. Setting the gradient of the objective function to 0, resulting in a corresponding closed solution:
The minimization problem of the function solved by the steepest descent method with the optimal step length can be expressed as follows:
whereinIs the gradient of the objective function, is the optimal step size.Taking into account the estimated valuesThe present embodiment applies the minimum mean square error theorem to change the problem of equation (25) into:
similarly, the question of (27) can be written as:
Due to the fact thatThe definition of (A) is complicated, and the above formula is difficult to be solved intuitively. This embodiment derives a set of closed solutions by proposing reasonable assumptions. Will be provided withOf a type of observation viewed as noisy, usingRepresenting the error phasor, the error of each element is respectively() Then assume e is independently distributed over Mean 0 and varianceFrom law of large numbers, for arbitraryAll have:
so that the following results are obtained:
the transformed error phasor is, Representing the error phasor for each element, and m is the number of data points for the best match. The transformation does not change the variance of each group according to the orthogonal nature of the matrix. It follows therefrom that of each groupAre all independently distributed in(zero mean and variance of) Most of them are theoretic and arbitraryAll have:
the method comprises the following steps:
it can be easily obtained that:
combining the problem with the formula (26) to obtain:
Under the vector contraction form, the estimated value of x is expressed as:
after the estimation value of the Taylor coefficient matrix x is obtained, the method is carried outA harmonic sampling equation for realizing the harmonic signalThe purpose of accurately detecting the harmonic phasor change is achieved. In order to ensure the accuracy of the model, under the framework of logistic regression, a cross entropy objective function is introduced to measure the probability distribution difference between an estimated value and a theoretical value x:
assuming that the error is a binary distribution,the predicted probability distribution can be considered to be very closely related to the actual probability distribution, which demonstrates that the assumption is consistent with the expected model. The algorithm flow is shown in the following table:
3. simulation analysis
This section discusses the result of the broadband dynamic phasor measurement algorithm (BMP) proposed in this embodiment under different test scenarios. The following 4 algorithms are used as comparison algorithms: fast fourier method (FFT), Prony algorithm, Taylor Weighted Least Squares (TWLS), dynamic synchrophasor measurement algorithm (SIFE) based on sinc interpolation function. The test scenario includes a basic performance test, an interference test, a frequency ramp test, and a step change test.
To compare the estimation effects of the different methods, a total phasor error (TVE) was introduced to describe the relative deviation between the theoretical phasor and the estimated phasor. The TVE is closely related to the amplitude error and the phase angle error, but cannot reflect the change of one aspect alone. Therefore, the present embodiment also introduces two indicators: the Frequency Error (FE) and the absolute frequency rate error (RFE) are used to evaluate the phasor estimation effect. In this embodiment, according to the IEEEC37.118.1 standard, the M-type PMU measurement requirement is selected as a standard limit value under different working conditions.
The 5 algorithms all use the same rectangular observation window, the bandwidth of the fundamental frequency is 1Hz, and the length of the sampling window is set as 5 power frequency periods. In order to improve the frequency domain resolution, the present embodiment considers the complexity of the algorithm and the accuracy of the algorithm together, and sets the value to 20.
4.1, basic Performance testing
To test the basic performance of the proposed algorithm, a broadband dynamic signal model containing DDC components is constructed as shown in equation (42):
wherein, the first and the second end of the pipe are connected with each other,for the fundamental frequency, here set at 50Hz,、respectively represent the fundamental wave and each harmonic phase angle, and are inAny value within the range. The low-frequency band harmonic times h are 2-13, the high-frequency band harmonic times h are 77, 79, 80 and 83 respectively, and the sampling frequency is set to be 10 kHz. DDC component magnitudeIs 0.3, 0.4, 0.5, … 1. Time constant of DDC componentStarting from 0.01s, it changes from 0.01s to 0.1s in steps.
Iterative phasor initial values when applying the BMP algorithmRegularization coefficient. Fixed value parameter,, When used, their zeta value is generally in a range [0.05,0.3 ]]An internal change. The number of data points in the search window that match best is set to m. In the process of matrix blocking, if the number of matched blocks is too large, data points with large noise influence and low matching degree inevitably exist in the block array; otherwise, the influence of the contingency on the construction matrix cannot be avoided. In this embodiment, the maximum iteration number is J, and the higher the iteration number is, the higher the calculation accuracy is, but the calculation cost is significantly increased. In this embodiment, the reconstruction effect and the algorithm running time are analyzed under the condition that the parameters m and J are changed, and the result is shown in fig. 1.
FIG. 1 is a graph of reconstruction effect and run-time range as a function of parameters. As can be seen from fig. 1, as the parameters m and J increase, the total phasor error shows a decreasing trend, but the algorithm runtime increases. When the number of data points m =240 and the number of iterations J is less than 100, changing the number of iterations has less influence on the calculation cost (the change in the running time is small) and the reconstruction effect tends to be stable at this time.In time, the increasing number of iterations leads to a great increase in computational complexity (large variation in running time), and the TVE also exhibits an unstable downward trend. Therefore, appropriately reducing m can effectively improve the reconstruction effect. Considering both reconstruction performance and runtime, the present embodiment applies the algorithm with the parameter m =240 and the iteration number J = 100.
FFT, Prony, TWLS and SIFE are selected as comparison algorithms, the total phasor error estimation and frequency change rate error estimation results of the comparison algorithms are shown in the following table, and the estimation precision of each algorithm is compared with that of the following table:
as can be seen from the above table, the maximum values of the TVE, FE and RFE indices of the BMP algorithm are 2.79%, 0.096Hz and 2.437Hz/s, respectively. When harmonic components are included, the IEEE standard limits for TVE and FE are 3% and 0.1Hz, respectively, and the RFE limit is 2.7 Hz/s. Compared with other comparison algorithms, the estimation index of the invention completely meets the requirements of the IEEE standard. The result shows that the method still has good detection effect under the condition of the DDC component-containing broadband harmonic wave, and the phasor estimation precision is highest. The BMP uses a sparse distribution of harmonic frequency domain distributions to identify the most relevant components in the signal, which significantly improves the accuracy of the measurement results.
The estimation error of the SIFE method in the vicinity of the low harmonic does not meet the measurement standard of IEEE. This is because the SIFE method is a low pass filter based on baseband signal filtering, and it is difficult to obtain a zero error result. But because the harmonic synchronous phasor can be well estimated due to the wide passband and the wide stopband, the performance of the invention is superior to FFT and TWLS. The FFT measurement results are most affected by spectral leakage, which greatly reduces the accuracy of harmonic parameter identification. The maximum total phasor error exceeds 8%, and the accuracy of the FE and RFE measurement results is not ideal. Under dynamic conditions, the fourier transform model cannot track phasor changes in the observation window, resulting in incorrect phasor evaluation. The TWLS method uses a second order taylor order to fit the signal components. However, the taylor signal model has a large error and a limited accuracy. Increasing the taylor model order can reduce the model error, but the higher the order, the worse the passband performance of the filter. The Prony algorithm uses a parametric model to calculate the signal parameters. However, the estimation order thereof will limit the number of estimated frequency components, and the frequency estimation error will gradually increase.
4.2 frequency ramp test
A power imbalance between the load and the generator may cause the frequency of the broadband signal to decrease with increasing load and increase with increasing input power. To analyze the performance of BMP (the algorithm of this example) in a frequency ramp, the signals provided are as follows:
wherein the content of the first and second substances,is the fundamental frequency and takes the value of 50 Hz.The value of the present embodiment is 1 Hz/s.Anda fundamental phase and a harmonic phase, respectively, the phases being set toRandom numbers uniformly distributed within the range. The harmonic phasor estimation, frequency estimation and frequency change rate estimation results of the present embodiment and the comparative algorithm are shown in fig. 1. Assume a sampling frequency of 10kHz and a sampling window length of 5 power frequency cycles. Analysis of the signal of (42) using FFT, Prony, TWLS, SIFE as a comparison algorithm, FIG. 2[ a ]]FIG. 2[ b ]]FIG. 2[ c ]]FIG. 2[ a ] is the estimation results generated by each method under the frequency ramp condition]FIG. 2[ b ]]FIG. 2[ c ]]FIG. 2[ a ] is a graph comparing the maximum error of TVE, FE, RFE under the condition of frequency ramp]FIG. 2[ b ] is a graph of maximum error versus TVE under frequency ramp conditions]FIG. 2[ c ] is a graph comparing the maximum error of FE under the frequency ramp condition]For RFE under frequency ramp conditionsMaximum error vs. plot.
As can be seen from fig. 2 a, 2 b, and 2 c, the maximum error of TVE in this embodiment is 1.06%, the maximum error of frequency is 0.0095Hz, and the maximum error of frequency conversion rate is 0.103Hz/s, which all satisfy the IEEE standard requirement [14 ]. The result shows that the method can still keep higher precision when the fundamental frequency changes greatly and the frequency changes linearly, and the estimation effect is superior to other four algorithms. The BMP algorithm adopts a Taylor order to approximate a dynamic signal model, can accurately estimate the frequency and the phase angle, and is minimally influenced by linear frequency change. The method achieves the highest phasor estimation accuracy. In other algorithms, the FFT method cannot track the frequency change in real time under dynamic conditions, and thus the frequency change rate is large. Since TWLS is an estimation algorithm based on a dynamic model, Prony algorithm can accurately extract the low-frequency oscillation characteristic value of the dominant mode, the error calculation results of Prony and TWLS algorithms are smaller than FFT, the error characteristics of parameters such as phase angle and frequency of the algorithms are slightly influenced by frequency change, but the measurement results still cannot meet the requirements of IEEE. The Prony algorithm has difficulty obtaining accurate mode parameters if the time-varying characteristics of the signal are not well understood. The STWLS also cannot accurately estimate the higher harmonic phasor due to the influence of the fundamental component.
4.3 step Change test
In order to simulate fault conditions of sudden changes in the amplitude and phase of the voltage/current signal, it is necessary to simulate the proposed algorithm under these conditions to evaluate the response time and delay. At the start of the test, the amplitude of each component was set to 115% of the initial amplitude, and the phase was changed to. The broadband dynamic signal provided is as follows:
the test in this section assumes a sampling rate of 5kHz and a sampling period of 5 cycles. In the standard, the performance of each algorithm under a step change condition is evaluated by the speed of the response time. It is defined as the time interval between the first and last instants greater than a given threshold. The threshold values for the maximum TVE, FE and RFE values were 1.5%, 0.13Hz, and 0.78Hz/s according to the IEEE Standard under the test conditions in this section. The specific simulation results are shown in fig. 3.
As can be seen from fig. 3, the BMP method requires less time to reach the TVE, FE and RFE values of the IEEE standard under the amplitude and phase step change conditions than other algorithms. It can be considered that under the condition of step change, the estimation precision of the invention is the highest and the response time completely meets the standard requirement of P-type PMU. For the evaluation of the entire fourier transform of the step-wise smooth function, the FFT algorithm requires very complex multiplication operations. TWLS performance can be improved by recalculating its coefficients in each report frame, however this greatly increases the computational burden, requiring the computation of the pseudo-inverse. In order to achieve the precision requirement of the SIFE algorithm, a complex matrix of 70 rows and 1400 columns needs to be stored in a memory, and the sub-actual multiplication and the sub-actual addition need to be processed in real time, so that the memory and the processing capacity are very limited. In order to obtain more accurate analysis results by the Prony algorithm, the model order of the method needs to be increased, so that the calculated amount is increased.
4.4, anti-interference test
In general, some inter-harmonics and noise are contained in the power system signal, which seriously affect the estimation of the harmonic phasor. In this section of the test, white gaussian noise with a signal to noise ratio of 55dB was introduced into the signal. Specific broadband dynamic signals are as follows:
wherein, the first and the second end of the pipe are connected with each other,is the inter-harmonic frequency, and the values thereof are 9652.5Hz, 9751.5Hz, 9850.5Hz, 9949.5Hz, 10048.5Hz and 10147.5Hz in sequence.The sum is the phase of the fundamental wave and the harmonic wave, and takes the value ofRandom numbers within a range. This section assumes a sampling frequency of 10kHz and a sampling window length of 5 power frequency cycles. The specific simulation results are shown in FIG. 4[ a ]]FIG. 4[ b ]]FIG. 4[ c ]]Shown in FIG. 4[ a ]]FIG. 4[ b ]]FIG. 4[ c ]]FIG. 4 is a graph of maximum error of TVE, FE and RFE under inter-harmonic interference [ a ]]FIG. 4[ b ] is a graph of the maximum error of TVE under inter-harmonic interference]FIG. 4[ c ] is a graph of the maximum error of FE under inter-harmonic interference]Is a maximum error map of RFE under inter-harmonic interference.
As can be seen from FIGS. 4[ a ], 4[ b ] and 4[ c ], the TVEmax of the algorithm of the present embodiment is 2.43%, FEmax is 0.071Hz, and RFEmax is 0.184 Hz/s. The calculation result of each index of the algorithm of the embodiment is superior to that of the other comparison algorithms. Under the condition of the existence of inter-harmonic interference, the IEEE standard limit values of TVE, FE and RFE are respectively 3.5%, 0.2Hz and 3 Hz/s. In the same way, the parameter results of the SIFE algorithms TVEmax, FEmax and RFEmax are respectively 6.64%, 0.115Hz and 7.39Hz/s, the Prony algorithms are 19.73%, 0.136Hz and 9.207Hz/s, the TWLS algorithms are 26.35%, 0.171Hz and 19.634Hz/s, and the FFT algorithms are 49.76%, 17.294Hz and 24.270Hz/s, and the calculation results of all indexes of the invention are superior to other comparison algorithms. Each index of the BMP algorithm meets the requirements of the IEEE standard. In the presence of noise interference in this segment, severe interference and spectral leakage can occur between adjacent harmonics of the FFT method, affecting resolution and accuracy. Under the condition of noise interference in the present segment, the TWLS and SIFE methods can increase the transition band amplitude of their harmonic filters, causing interference between adjacent harmonics, resulting in larger estimation error. In addition, the TWLS algorithm is severely affected by inter-harmonic interference, and it is difficult to accurately estimate each high-frequency component. For high-order components, the model parameters of the Prony algorithm are continuously modified along with the increase of the harmonic order. Therefore, it has a good effect in inter-harmonic phasor and frequency estimation, and can suppress the influence of the inter-harmonic component spectrum leakage to some extent. However, this method requires the order of the dynamic time-varying signal to be estimated in advance.
The BMP algorithm adopts an improved Taylor-Fourier model and a machine learning algorithm, accurately recovers a specific signal by fewer data points, reconstructs harmonic phasors and solves the specific signal by a sparse acquisition model, and effectively improves the reconstruction performance and the anti-noise capability of the algorithm. In addition, the spectral resolution of the reconstruction of the invention is 1Hz, which is beneficial to the accurate detection of the inter-harmonic component.
4. Conclusion
The harmonic detection method based on the sparse acquisition model provided by the embodiment is used for solving the problem of effective processing of the broadband dynamic phasor containing the DDC component. The phasor estimation method is based on the idea of regularizing the sparse capture matrix, and obtains phasor estimation results by solving a sparse regularization problem. Simulation and experiment results show that the method identifies key information of the broadband dynamic phasor, and obviously reduces the computational complexity. This represents the ability to obtain more accurate results in a short time, thereby effectively detecting the transient characteristics of a broadband signal. Meanwhile, under static and dynamic conditions such as noise interference, inter-harmonic interference, frequency slope and the like, the measurement result can meet the test requirements of the M-level PMU.
The above-mentioned embodiments only express the specific embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (10)
1. A harmonic detection method based on a sparse acquisition model is characterized by comprising the following steps:
a. estimating the attenuation direct current component in the broadband dynamic signal: establishing a broadband dynamic signal model, and calculating an attenuated direct current component;
b. estimating harmonic phasor in the broadband dynamic signal: establishing a harmonic phasor estimation model, reconstructing the harmonic phasor estimation model, and solving the harmonic estimation model;
c. a simulation analysis step: and establishing a test scene, and sequentially establishing a basic performance test scene, a frequency slope test scene, a step change test scene and an interference test scene.
2. The harmonic detection method based on the sparse acquisition model as claimed in claim 1, wherein in the step a, the step of establishing the broadband dynamic signal model is as follows:
for power system waveforms containing a DC component and a fundamental componentThe expression is shown as (1):
wherein the content of the first and second substances,is a signal containing fundamental, harmonic and inter-harmonic components,is a time-varying amplitude of the signal,is a time-varying phase angle, the reference frequency is,Refers to the highest order harmonic component of the signal,is to attenuate the signal of the direct current component,is the amplitude of the attenuated dc component,is a time constant;
signals containing fundamental, harmonic and inter-harmonic componentsThe dynamic phasor model at time t is:
substituting equation (2) into equation (1) to obtain broadband dynamic signalThe dynamic expression of (2):
3. The harmonic detection method based on the sparse acquisition model as claimed in claim 2, wherein in the step a, the step of calculating the attenuated direct current component is as follows:
attenuating DC componentRepresented by the sum of the amplitude and phase time-varying cosine components:
wherein, the first and the second end of the pipe are connected with each other,,,respectively representing the frequency, the amplitude and the phase of the nth cosine component, wherein T is the length of an observation interval, and Y is set to be 3;
based on frequency domain sampling theorem, a sine interpolation function is utilized to carry out time domain signal parameterization modeling on the attenuation direct current component, and the dynamic phasor of the DDC component is recorded asSpecifically, it is represented as:
wherein, the first and the second end of the pipe are connected with each other,is the frequency of the sampling of the samples,indicating the maximum model order, will be set to 2,is the nth cosine component in the formula (2) at the timeA sample phasor above;
for the attenuation of DC componentSampling is carried out, and the discrete form of the fitted DDC component in the formula (1) is represented by (6);
wherein, the first and the second end of the pipe are connected with each other,representing phasorsIs a column vector of the cosine signal,as conjugate operators, matricesIs an element ofAt a sample point in the sampling window(s),representation matrix,The definition of (2) is shown as (7):
determining phasor estimation values by means of least squaresThe calculation process is shown in (7) in the matrixIn (1),for the matrix elements:
where H is a conjugate transposed symbol;
4. The harmonic detection method based on the sparse acquisition model as claimed in claim 1, wherein in the step b, the step of establishing the harmonic phasor estimation model is as follows:
the dynamic phasors are defined using a taylor expansion model of order:
wherein the content of the first and second substances,is t =0K is the taylor expansion order, and T is the observation interval length;
the sinusoidal signals containing harmonics and inter-harmonics take the form of discrete sequencesThe sequence length is N, andand expanding the dynamic phasor method to an actual sequence model through multi-frequency phasor analysis, wherein the Taylor Fourier coefficient expression of each phasor component of the discrete signal is as follows:
where H is the sampling interval, observation interval length,Is the average harmonic and inter-harmonic phasors,is a phasor frequency ofThe derivative of the order k of (c) is,is the fundamental frequencyInteger multiples of (d) represent harmonic frequencies, and non-integer multiples represent inter-harmonic frequencies;
in the size ofThe sampling rate of (2) normalizes the frequency of the harmonic signal by a sampling length ofTo obtain the frequency of harmonic componentThe normalized frequency of the Taylor Fourier basis vector is,The frequency resolution corresponding to the N coefficient sets isAnd the taylor opening order of each i is K, then (10) is written in matrix form:
wherein, the first and the second end of the pipe are connected with each other,is a sample column phasor of sizeIs a Taylor Fourier-based phasor, e represents noise or measurement error, and x is a length ofThe column phasors of (1), describe the current harmonicsWhen the temperature of the water is higher than the set temperature,a set of (a);
and (3) phasor solving, namely converting the phasor solving into solving of a pseudo-inverse matrix by using a least square method, and calculating the pseudo-inverse matrix to obtain a corresponding solution when the Euclidean norm is minimum, wherein the solution is the optimal solution under the following constraint conditions:
wherein the content of the first and second substances,representing Euclidean norm, and further obtaining a dynamic phasor coefficient estimation matrix after introducing a Lagrange operator and derivation:
5. The harmonic detection method based on the sparse acquisition model as claimed in claim 4, wherein in the step b, the step of reconstructing the harmonic phasor estimation model is as follows:
setting a correction factorEach group of frequency resolution reachesThe length of the sampling sequence isThe sampling length H is 5 fundamental wave periods, and the harmonic component normalized frequency is,And obtaining an improved Taylor Fourier coefficient expression:
wherein, the first and the second end of the pipe are connected with each other,,is expressed as a size ofThe matrix of the curvelet transform of (c),is thatThe k-th derivative of (a);
according to the expression (9), an approximate second-order dynamic phasor estimated value and harmonic frequency are obtainedAnd rate of change of frequency:
When the correction factor isWhen the sensing matrix A and the sparse matrix are in useHighly uncorrelated, (14) is converted into matrix form, namely:
wherein, the first and the second end of the pipe are connected with each other,is a measurement matrix, x represents a length ofThe corresponding sample sequence of each data block of (a),is the superposition of a small amount of phasors extracted after the original sample is segmented,is dimension ofEach column of the matrix is a base phasor of a discrete fourier transform, e represents a noise signal phasor;
after the sparse sampling measurement model is constructed by the method, matrix blocking is carried out on x by utilizing an Euclidean search algorithm, m data points which are optimally matched are found in a search range, and all the data points are combined into one matrixI is a coordinate of the coefficient matrix, curvelet transformation is carried out on the matrix, an insignificant curvelet coefficient is removed by adopting a curvelet threshold criterion, an effective denoising and compression algorithm is required for the extracted coefficient, curvelet transformation is carried out, an insignificant curvelet coefficient is removed by adopting a curvelet threshold criterion, and an effective denoising and compression algorithm is required for the extracted coefficient;
sparse matrixOnly contains a non-zero significant term set, the transformation coefficient column phasor obtained after the significant term set is arranged according to the dictionary order isTransformation coefficientIs focused onThe distribution is near the zero zone and in the form of a fine peak, and the repeatability of the signal is represented by using the distribution characteristics of the transformation coefficient, and is recorded as follows:
wherein, the first and the second end of the pipe are connected with each other,represents the L1 norm;
each set of harmonic signal estimates may be inverted according to transform coefficients:
6. The harmonic detection method based on the sparse acquisition model as claimed in claim 5, wherein in the step b, the step of solving the harmonic estimation model is as follows:
solving a regularized linear least square cost function, and reconstructing a Taylor coefficient matrix x, namely:
wherein the content of the first and second substances,the norm of L2 is shown,a regularization term is represented as a function of,is a function of the regularization coefficients and,、representing local and non-local sparse terms respectively,is a regularization parameter that balances the two sparse terms;
(22) substituting (21) to obtain:
introducing two auxiliary phasors p and q, finally the following scheme is obtained:
wherein the content of the first and second substances,andthe parameters are fixed value parameters, the function is to improve the stability of the algorithm value, and p and q are SBI algorithm auxiliary iteration phasors;
(24) solving the minimization of a strict convex quadratic function, setting the gradient of an objective function as 0, and obtaining a corresponding closed solution:
wherein, the first and the second end of the pipe are connected with each other,is of the sizeThe identity matrix of (a);
the minimization problem of the function is solved by adopting the steepest descent method with the optimal step length, which is expressed as follows:
wherein the content of the first and second substances,is the gradient of the objective function and,is the optimum step size for the particular application,the estimated value isApplying the least mean square error theorem changes the problem of (25) into:
similarly, the question of (26) is written as:
Obtaining a set of closed solutions, willObserved values seen as noiseOf a type usingRepresenting the error phasor, the error of each element is respectively() Then assume e is independently distributed overMean 0 and varianceFrom law of large numbers, for arbitraryAll have:
so that the following results are obtained:
the transformed error phasor is, The error phasor representing the error phasor for each element,m is the number of data points of the best match, and the variance of each group is not changed by transformation according to the orthogonal property of the matrix, thereby obtaining the data of each groupAre all independently distributed in(zero mean and variance of) Most of them are theoretic, for anyAll have:
the method comprises the following steps:
to obtain:
combining the two problems with the (26) problem to obtain:
wherein the content of the first and second substances,refers to the modulus of the corresponding phasor;
under the vector contraction form, the estimated value of x is expressed as:
after the estimation value of the Taylor coefficient matrix x is obtained, the method comprises the following stepsHarmonic sampling equation to harmonic signalsIn the framework of logistic regression, cross entropy was introduced [27 ]]The objective function measures the difference in probability distribution between the estimated and theoretical values x:
7. The harmonic detection method based on the sparse acquisition model as claimed in claim 1, wherein in the step c, the step of establishing a basic performance test scenario is as follows:
the basic performance of the proposed algorithm is tested, and a broadband dynamic signal model containing DDC components shown in formula (43) is constructed:
wherein the content of the first and second substances,which is the fundamental frequency, here set at 50Hz,、respectively represent the fundamental wave and each harmonic phase angle, and are inThe value of the harmonic frequency h of the low frequency band is 2-13, the harmonic frequency h of the high frequency band is 77, 79, 80 and 83 respectively, the sampling frequency is set to 10kHz, and the amplitude of DDC component is set to be any value in the rangeIs taken to be 0.3, 0.4, 0.5, … 1, the time constant of DDC componentStarting from 0.01s, changing from 0.01s to 0.1s in steps;
8. The harmonic detection method based on the sparse acquisition model as claimed in claim 7, wherein in the step c, the step of establishing a frequency ramp test scenario is as follows:
the BMP was analyzed for performance under a frequency ramp, providing the following signals:
9. The harmonic detection method based on the sparse acquisition model as claimed in claim 8, wherein in step c, the step of establishing the step change test scenario is as follows:
at the start of the test, the amplitude of each component was set to 115% of the initial amplitude, and the phase was changed toThe broadband dynamic signal provided is as follows:
the sampling rate is set to be 5kHz, the length of the sampling period is 5 periods, and in the standard, the performance of each algorithm under the condition of step change is evaluated by the speed of response time.
10. The harmonic detection method based on the sparse acquisition model as claimed in claim 9, wherein in the step c, the step of establishing an interference test scenario is as follows:
gaussian white noise with the signal-to-noise ratio of 55dB is introduced into the signal, and specific broadband dynamic signals are as follows:
wherein, the first and the second end of the pipe are connected with each other,is the inter-harmonic frequency, and the values thereof are 9652.5Hz, 9751.5Hz, 9850.5Hz, 9949.5Hz, 10048.5Hz and 10147.5Hz in sequence,andis fundamental wave and harmonic waveAnd takes on a value ofThe random number in the range is set to have a sampling frequency of 10kHz and a sampling window length of 5 power frequency periods.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210698874.3A CN114781196A (en) | 2022-06-20 | 2022-06-20 | Harmonic detection method based on sparse acquisition model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210698874.3A CN114781196A (en) | 2022-06-20 | 2022-06-20 | Harmonic detection method based on sparse acquisition model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114781196A true CN114781196A (en) | 2022-07-22 |
Family
ID=82421455
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210698874.3A Pending CN114781196A (en) | 2022-06-20 | 2022-06-20 | Harmonic detection method based on sparse acquisition model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114781196A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115032453A (en) * | 2022-08-08 | 2022-09-09 | 四川大学 | Multi-frequency dynamic phasor measurement method |
CN115343532A (en) * | 2022-08-09 | 2022-11-15 | 国网福建省电力有限公司 | Method for compressing and reconstructing disturbance-containing power quality signal based on compressed sensing |
CN115856767A (en) * | 2023-02-08 | 2023-03-28 | 安徽大学 | Reconfigurable intelligent super-surface-assisted wave arrival direction estimation method |
CN116150594A (en) * | 2023-04-18 | 2023-05-23 | 长鹰恒容电磁科技(成都)有限公司 | Method for identifying switch element characteristics in spectrum test data |
CN117200242A (en) * | 2023-11-08 | 2023-12-08 | 西安米特电子科技有限公司 | Monitoring data processing method and system for intelligent voltage regulating cabinet |
CN117639876A (en) * | 2024-01-25 | 2024-03-01 | 南京理工大学 | Linear frequency modulation wave anti-interference DOA estimation method based on space-time modulation super surface |
-
2022
- 2022-06-20 CN CN202210698874.3A patent/CN114781196A/en active Pending
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115032453A (en) * | 2022-08-08 | 2022-09-09 | 四川大学 | Multi-frequency dynamic phasor measurement method |
CN115343532A (en) * | 2022-08-09 | 2022-11-15 | 国网福建省电力有限公司 | Method for compressing and reconstructing disturbance-containing power quality signal based on compressed sensing |
CN115856767A (en) * | 2023-02-08 | 2023-03-28 | 安徽大学 | Reconfigurable intelligent super-surface-assisted wave arrival direction estimation method |
CN116150594A (en) * | 2023-04-18 | 2023-05-23 | 长鹰恒容电磁科技(成都)有限公司 | Method for identifying switch element characteristics in spectrum test data |
CN116150594B (en) * | 2023-04-18 | 2023-07-07 | 长鹰恒容电磁科技(成都)有限公司 | Method for identifying switch element characteristics in spectrum test data |
CN117200242A (en) * | 2023-11-08 | 2023-12-08 | 西安米特电子科技有限公司 | Monitoring data processing method and system for intelligent voltage regulating cabinet |
CN117200242B (en) * | 2023-11-08 | 2024-02-02 | 西安米特电子科技有限公司 | Monitoring data processing method and system for intelligent voltage regulating cabinet |
CN117639876A (en) * | 2024-01-25 | 2024-03-01 | 南京理工大学 | Linear frequency modulation wave anti-interference DOA estimation method based on space-time modulation super surface |
CN117639876B (en) * | 2024-01-25 | 2024-04-23 | 南京理工大学 | Linear frequency modulation wave anti-interference DOA estimation method based on space-time modulation super surface |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114781196A (en) | Harmonic detection method based on sparse acquisition model | |
CN108037361B (en) | High-precision harmonic parameter estimation method based on sliding window DFT | |
Boashash et al. | Use of the cross Wigner-Ville distribution for estimation of instantaneous frequency | |
Messina et al. | Nonstationary approaches to trend identification and denoising of measured power system oscillations | |
Yu | A multisynchrosqueezing-based high-resolution time-frequency analysis tool for the analysis of non-stationary signals | |
CN107102255B (en) | Single ADC acquisition channel dynamic characteristic test method | |
CN111289796B (en) | Method for detecting subsynchronous oscillation of high-proportion renewable energy power system | |
Pan et al. | A noise reduction method of symplectic singular mode decomposition based on Lagrange multiplier | |
CN110311392B (en) | Electric power system oscillation mode and mode identification method based on SPDMD | |
CN109659957B (en) | APIT-MEMD-based power system low-frequency oscillation mode identification method | |
Karpilow et al. | Characterization of non-stationary signals in electric grids: A functional dictionary approach | |
Philip et al. | An improved Stochastic Subspace Identification based estimation of low frequency modes in power system using synchrophasors | |
JP2014153354A (en) | Method for estimating frequencies and phases in three phase power system | |
Mei et al. | Wavelet packet transform and improved complete ensemble empirical mode decomposition with adaptive noise based power quality disturbance detection | |
CN112816779B (en) | Harmonic real signal parameter estimation method for analytic signal generation | |
CN112883318A (en) | Multi-frequency attenuation signal parameter estimation algorithm of subtraction strategy | |
Bertocco et al. | Frequency tracking for efficient phasor measurement based on a CSTFM model | |
CN115032453A (en) | Multi-frequency dynamic phasor measurement method | |
CN109254202B (en) | Synchronous phasor measurement device applied to power distribution network | |
CN114184838A (en) | Power system harmonic detection method, system and medium based on SN mutual convolution window | |
Hu et al. | Aerodynamic loads correction method based on wavelet packet transform in high-frequency force balance tests | |
Xingang et al. | Supraharmonics measurement algorithm based on CS-SAMP | |
CN111398735A (en) | Transformer substation grounding grid fault detection method based on information entropy | |
Frigo et al. | Compressive sensing Taylor-Fourier and windowing approach for synchronized phasor, frequency and ROCOF measurements | |
CN116522269B (en) | Fault diagnosis method based on Lp norm non-stationary signal sparse reconstruction |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220722 |
|
RJ01 | Rejection of invention patent application after publication |