CN113127801A - Electric power system oscillation parameter identification method, system, terminal and storage medium - Google Patents

Electric power system oscillation parameter identification method, system, terminal and storage medium Download PDF

Info

Publication number
CN113127801A
CN113127801A CN202110411632.7A CN202110411632A CN113127801A CN 113127801 A CN113127801 A CN 113127801A CN 202110411632 A CN202110411632 A CN 202110411632A CN 113127801 A CN113127801 A CN 113127801A
Authority
CN
China
Prior art keywords
oscillation
signals
signal
frequency
band
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
Application number
CN202110411632.7A
Other languages
Chinese (zh)
Inventor
孙琳
高萍
张馨元
洪宁
曹枫林
晏磊
朱小红
黄蓉蓉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuchang University of Technology
Original Assignee
Wuchang University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuchang University of Technology filed Critical Wuchang University of Technology
Priority to CN202110411632.7A priority Critical patent/CN113127801A/en
Publication of CN113127801A publication Critical patent/CN113127801A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/002—Flicker reduction, e.g. compensation of flicker introduced by non-linear load
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]

Abstract

The invention belongs to the technical field of power systems, and discloses a method, a system, a terminal and a storage medium for identifying oscillation parameters of a power system, wherein signals of different frequency bands are separated through a band-pass filter; extracting oscillation mode signals in each band-pass filtering signal by using an Improved Variational Mode Decomposition (IVMD) method; the HTLS algorithm and the adaptive artificial neural network algorithm Adaline ANN are combined, the frequency, the attenuation factor, the amplitude and the phase of the low-frequency oscillation are respectively estimated, and the universal frequency band oscillation is uniformly identified. The invention can effectively carry out oscillation distinguishing, mode extraction and parameter identification on multi-type and universal-frequency-band oscillations such as low-frequency oscillation, subsynchronous oscillation, super-synchronous oscillation and the like from the operation data of the power system. The identification result plays an important role in the aspects of system dynamic stability analysis and control, oscillation source identification and positioning and the like, and is beneficial to early warning and timely inhibiting system oscillation.

Description

Electric power system oscillation parameter identification method, system, terminal and storage medium
Technical Field
The invention belongs to the technical field of power systems, and particularly relates to a method, a system, a terminal and a storage medium for identifying oscillation parameters of a power system.
Background
At present, with the large-scale access of renewable energy sources, the wide application of power electronic equipment and the large-scale interconnection of alternating current and direct current, the modern power grid optimizes resource allocation and improves system reliability, meanwhile, the number of weak links in the system is increased, the anti-interference performance is reduced, new types of faults occur, and the problem of safety and stability is obvious. Oscillation is one of the problems threatening the stable operation of the power system, and typical oscillations common to the current power system include: low Frequency Oscillation (LFO) induced by zone or interval weak damping effects and subsynchronous Oscillation (SSO) induced by series compensation capacitors or energy interaction between power electronics and gensets. In addition, Super-synchronous Oscillation (SurSO) occasionally occurs along with subsynchronous Oscillation. If the oscillation is not suppressed in time, the instability and even disconnection of the power grid can be caused.
Depending on the damping, the oscillation signals of the power system can be divided into two categories: (1) sustained or divergent oscillating signal (when the system is weakly or negatively damped); (2) the oscillating signal is damped (when the system is in positive damping). The former has low occurrence frequency, but causes great harm to the system once occurring; the latter occur more frequently and are easily masked by noise, also known as noise-like oscillation signals. Modal information is quickly and timely identified from the oscillation signals, so that the scheduling of a power grid and the adjustment of a control strategy are facilitated; before oscillation occurs, mode identification is carried out on the noise-like oscillation signals, and potential oscillation modes of the system can be found, so that early warning is provided for the system. The oscillation mode identification is the key for the real-time efficient control and risk early warning of the power system.
The frequency, the damping ratio, the oscillation amplitude and the phase are key information of an oscillation mode and are key parameters for realizing oscillation monitoring, early warning, control and protection research. At present, the research method of the oscillation of the power system mainly comprises model analysis and measurement analysis. The model analysis method depends on an accurate model of a system, and large systems and nonlinear power electronic devices can influence the model accuracy, so that analysis errors are caused. The measurement analysis is to analyze the actual measurement data of the system, and the analysis result is closer to the operation condition. The combination of the measured data and the signal processing technology is a common method for identifying the oscillation of the power system at present.
The Prony algorithm can directly extract the characteristic quantity of the signal, so that more accurate oscillation mode information can be identified, but the Prony algorithm has high noise sensitivity. In practical engineering, a filtering algorithm is often combined with a Prony algorithm, and the filtering algorithm is used for reducing noise of a signal or extracting a modal signal from the signal. Methods such as Average Filter (AF), empirical mode decomposition, auto-regressive moving Average algorithm, wavelet method, singular value decomposition, etc. are commonly used filtering algorithms and have been applied to oscillation mode identification. However, the mean filtering can only reduce the interference of noise and cannot filter out the noise; the essence of wavelet denoising is that a signal is fitted in a specific frequency band, and if a plurality of oscillation modes with similar frequencies exist, a wavelet method cannot distinguish the signals; the hilbert transform, singular value decomposition, and other methods may have a large error when processing low signal-to-noise ratio signals. In addition, in the face of the problem of wide-band oscillation with multiple concurrent types and large oscillation frequency span, the method often identifies the oscillation with higher amplitude in the processing process, and other oscillations can be considered as noise and ignored. How to perform unified mode identification on the pan-band oscillation signals and extract multi-class oscillation mode information is an important problem faced by the current oscillation mode identification of the power system. The Variational Mode Decomposition (VMD) algorithm can effectively separate modal signals and is insensitive to noise.
Through the above analysis, the problems and defects of the prior art are as follows:
(1) the model analysis method depends on an accurate model of a system, and large-scale systems and nonlinear power electronic devices can influence the model accuracy, so that analysis errors are caused.
(2) The Prony algorithm has high sensitivity to noise; the mean filtering can only reduce the interference of noise but can not filter the noise; the essence of wavelet denoising is that a signal is fitted in a specific frequency band, and if a plurality of oscillation modes with close frequencies exist, a wavelet method cannot distinguish the signals; the hilbert transform, singular value decomposition, and other methods may have large errors when processing low signal-to-noise ratio signals.
(3) The prior method often identifies the oscillation with higher amplitude in the processing process, and other oscillations can be considered as noise and ignored.
The difficulty in solving the above problems and defects is:
in the parameter identification of the oscillation signal of the power system, the accuracy of the parameter identification is greatly influenced by mode aliasing and noise interference. Classic parameter identification methods such as window Fourier transform, wavelet transform, Hilbert transform, total minimum two-component rotation invariant technology (TLS-ESPRIT) and the like cannot overcome the influence of modal aliasing, false modal components are easily introduced in the identification process, and some unexplained frequency parameters are generated; although the autoregressive spectrum estimation method, the Prony algorithm, the neural network method and the like have certain improvement on mode aliasing, the autoregressive spectrum estimation method is sensitive to noise, and can generate wrong parameter estimation under noise interference, thereby causing serious consequences. How to simultaneously overcome the defects of the existing method and accurately extract the parameters of the oscillation signals under the condition of mode mixing and noise interference has very high difficulty and challenge.
The significance of solving the problems and the defects is as follows: oscillation is one of the problems threatening the stable operation of the power system, and typical oscillations common to the current power system include: the low-frequency oscillation caused by the weak damping effect of the region or the interval, the subsynchronous oscillation and the supersynchronous oscillation caused by the energy interaction between series compensation capacitors or power electronic equipment and a generator set.
Oscillation signals are often mixed together and are covered by noise, and if oscillation parameters cannot be accurately identified from the mixed and noisy signals or all oscillation modes cannot be completely extracted, oscillation of the power system cannot be timely inhibited, so that overcurrent tripping of a connecting line or desynchronization between systems or between units and systems are caused to cause disconnection, and the stability of the power system is seriously threatened.
Therefore, the method for rapidly and accurately identifying the parameters such as the frequency, the damping ratio, the oscillation amplitude, the phase and the like from the noise-containing and vibration-intensive oscillation signals is the key for the real-time and efficient control and risk early warning of the power system, and has important significance.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method, a system, a terminal and a storage medium for identifying oscillation parameters of a power system.
The invention is realized in such a way that the method for identifying the oscillation parameters of the power system comprises the following steps:
the separation of signals of different frequency bands is realized through a band-pass filter; extracting oscillation mode signals in each band-pass filtering signal by using an Improved Variational Mode Decomposition (IVMD) method; the HTLS algorithm and the adaptive artificial neural network algorithm Adaline ANN are combined, the frequency, the attenuation factor, the amplitude and the phase of the low-frequency oscillation are respectively estimated, and the universal band oscillation is uniformly identified.
Further, the method for identifying the oscillation parameters of the power system comprises the following steps:
using a plurality of band-pass filters (BPFs) to decompose an original signal into a plurality of filtered signals of different frequency bands, so as to realize the separation of oscillation signals of different frequency bands;
extracting oscillation mode signals from the BPF filtered signals by using an IVMD algorithm;
and step three, identifying oscillation mode signals provided by the IVMD through an HTLS-Adaline algorithm, and acquiring information of each oscillation mode.
Further, in the first step, the oscillation includes oscillation mode signals of a plurality of different frequency bands and noise signals generated by measurement or existing in the system; wherein the raw measurement signal y containing the pan-band oscillation0(t) is expressed as:
Figure BDA0003024392830000041
in the formula, yNoise(t) is a noise signal; y isn(t) (N is 1,2, …, N) is a modal signal of N different frequency bands, and:
Figure BDA0003024392830000042
in the formula, An(t) is the amplitude of the nth mode of oscillation at time t, θnIs the initial phase; dnAnd fnRespectively damping ratio and oscillation frequency of the oscillation.
Further, in the first step, the band pass filters are designed to be 4, and the BPF filtering bandwidth and the identification sampling duration corresponding to each type of oscillation include:
firstly, a band-pass filter BPF-1 is adopted, the oscillation type is an LFO interval mode, the sampling time length is 10s, and the oscillation frequency band is 0.2-1 Hz;
secondly, a band-pass filter BPF-2 is adopted, the oscillation type is an LFO local mode, the sampling time length is 2s, and the oscillation frequency band is 1-5 Hz;
thirdly, a band-pass filter BPF-3 is adopted, the oscillation type is SSO, the sampling time is 0.4s, and the oscillation frequency band is 10-50 Hz;
and fourthly, a band-pass filter BPF-4, wherein the oscillation type is SurSO, the sampling time is 0.4s, and the oscillation frequency band is 70-110 Hz.
Taking the lowest band-pass frequency of the band-pass filter of 0.2Hz as a reference, and selecting the sampling duration to be 2 times of the oscillation period, namely 10 s; considering the subsynchronous oscillation, the comprehensive precision and the identification speed, and taking the lowest oscillation frequency of 10Hz as a reference, wherein the sampling duration is 4 times of the oscillation period, namely 0.4 s; the mechanism of the super-synchronous oscillation determines that the super-synchronous oscillation and the sub-synchronous oscillation occur in pairs and have complementary frequencies, so that the sampling time of the super-synchronous oscillation is consistent with that of the sub-synchronous oscillation and is also 0.4 s.
Further, in step two, the modal signal extraction based on the improved VMD algorithm includes:
the variational problem corresponding to the VMD algorithm is to seek the K IMF components u with the minimum sum of the estimation bandwidthsk(t) of (d). Transforming the variation problem to obtain an augmented Lagrange equation, and solving the equation by adopting a multiplier Alternating Direction Multiplier Method (ADMM) to obtain a modal function
Figure BDA0003024392830000051
Solving the value problem:
Figure BDA0003024392830000052
similarly, the solution to the problem of the center frequency of the modal component is:
Figure BDA0003024392830000053
wherein, { u [ [ u ] ]k}＝{u1,u2,…,uK}，{ωk}＝{ω1,ω2,…,ωKThe components and their center frequencies.
The algorithm flow of the VMD is as follows:
(1) initialization
Figure BDA0003024392830000054
And n is 0;
(2) n ← n +1, and u is updated from expressions (1) and (2)kAnd ωk；
(3) Updating lambda:
Figure BDA0003024392830000055
(4) repeating steps (2) and (3) until an iteration stop condition is satisfied, i.e.
Figure BDA0003024392830000061
And ending circulation, and outputting a result to obtain K modal components and the center frequency thereof.
Further, in step two, the modal signal extraction based on the improved VMD algorithm further includes: and determining a VMD penalty factor based on PSO algorithm optimization.
The penalty factor alpha in the VMD algorithm has a large influence on the decomposition result, and researches show that: the smaller the penalty parameter alpha is, the larger the bandwidth of each obtained IMF component is, and conversely, the larger the alpha is, the smaller the bandwidth of each component is. Therefore, when the VMD is used to decompose the oscillation signal of the power system, it is very important to select an appropriate penalty factor parameter α. And optimizing the penalty parameters by adopting a genetic variation abnormal particle group algorithm to obtain an optimal alpha value.
Introducing the idea of genetic algorithm variation into the particle swarm algorithm to construct the genetic variation particle swarm algorithm.
Defining a genetic variation particle swarm algorithm: in a search space of D dimension, the marker population is X, and the species X is composed of m particles, i.e., X ═ X1,x2,…,xm]The position of each particle in the search space can be represented by a D-dimensional vector, that is:
xi＝[xi1,xi2,…,xiD]d is the number of parameters to be optimized; moving speed v of ith particlei＝[vi1,vi2,…,viD]Local extremum p of the particlei＝[pi1,pi2,…,piD]The generation groupGlobal extreme value G of1＝[g1,g2,…,gDSub-global optimum G2＝[g′1,g′2,…,g′D′]The maximum optimal retention algebra of an individual is maxAge, and the mutation probability is q. In order to prevent the particles from falling into local optimality, the individually optimal kept algebra of the particles in the iterative process needs to be recorded, when the individually optimal kept algebra does not reach maxAge, each particle updates the position and the speed of the next generation through an individual local extremum and a global extremum, and the updating formula is as follows:
Figure BDA0003024392830000062
Figure BDA0003024392830000063
where ω is the inertial weight and η is [0, 1 ]]Random number in between, c1And c2For learning factors, respectively representing local search capability and global search capability, n represents iteration number, v represents iteration numberi，pi，G1，xiAre all D-dimensional vectors. The determination of the inertia weight omega of the current iteration times adopts a linear decreasing weight method proposed by Shi, and the formula is as follows:
ω＝ωmax-(ωmax-ωmin)n/nmax
in the formula, ωmaxAnd ωminRespectively, the maximum and minimum inertia weight, n is the current iteration number, nmaxIs a defined maximum number of iterations. When the individual optimum kept algebra to reach maxAge, the position and speed of the particle are updated by genetic variation operation to make the particle jump out of local optimum.
Selecting a fitness function of the genetic variation particle swarm algorithm: during parameter optimization, the evaluation standard of the decomposition effect of the VMD method adopts the envelope entropy E proposed by Tang Guiji et alpNotion, the envelope entropy of a time signal x (j) of length N is defined as:
Figure BDA0003024392830000071
where a (j) is x (j) the envelope signal demodulated by hillbert, and j is 1,2, …, N. p is a radical ofjIs the result of normalizing a (j), EpIs obtained according to an information entropy calculation rule according to EpThe decomposition effect of the VMD is measured.
After the BPF-filtered oscillation signals are decomposed by a VMD method, if more noise is contained in the obtained components, the sparsity of the component signals is weaker, and the envelope entropy is larger; on the contrary, if regular oscillation signals appear in the components, the signals will present strong sparsity, and the calculated envelope entropy is small. Therefore, under the influence of the parameter alpha, the minimum envelope entropy E of the K components is selectedpAs local minimum entropy minEpThe component corresponding to the minimum entropy value contains abundant characteristic information. Using local minimum entropy as a part of fitness function in the whole search process, and searching a parameter alpha corresponding to the global optimum component0. Through the analysis of the parameter alpha, the number of iterations of the VMD is reduced by properly selecting alpha, namely the decomposition efficiency of the VMD method is higher. At minEpOn the basis, adding iteration time to construct a fitness function as follows:
minF＝minEp+β·time；
in the formula, β is a quantization factor of the fitness function, and β is 1/1000.
The VMD extracted oscillation mode u can be obtained by providing signal decomposition for BPF through the improved VMDn(t) of (d). For modal signal u when BPF and VMD accuracy is high enoughn(t) having:
un(t)≈yn(t)；
therefore, all oscillation mode signals can be separated from the original signals through the BPF and the IVMD.
Further, in step three, the solving of the frequency and the attenuation factor by the IVMD-HTLS algorithm includes:
(1) adaline neural network solution for amplitude and phase
In adaptive linear neural networks, x1k,x2k,…,xnkN input signals at time k for the adaptive linear neural network. The vector form of the input signal is represented as Xik＝[x1k,x2k,…,xnk]TOften referred to as the input pattern vector of the Adaline neural network. The weight vector corresponding to each group of input signals is Wik＝[w1k,w2k,…,wnk]T. The Adaline neural network output is:
Figure BDA0003024392830000081
let y (k) be the ideal response signal, defining the error function as:
Figure BDA0003024392830000082
the Adaline neural network works as follows: the ideal response signal y (k) is compared with the output signal of the neural network
Figure BDA0003024392830000083
Comparing to obtain difference e (k), transmitting e (k) to learning rule, adjusting weight vector according to learning rule, and making
Figure BDA0003024392830000084
In agreement with y (k).
The learning rule of the Adaline neural network is a Widrow-Hoff rule, namely a least mean square deviation algorithm LMS. The rule weight vector adjustment expression is:
Wi(k+1)＝Wi(k+1)+ηe(k)Xi(k+1)；
in the formula, η is the learning rate of the Adaline neural network, and η belongs to (0,1), and the value directly influences the weight vector regulation precision and the convergence speed.
(2) Adaline neural network solution for oscillation modes
The detailed steps for the Adaline neural network to solve for amplitude and phase are as follows. The known low-frequency oscillation discrete sampling signal model is:
Figure BDA0003024392830000085
when the attenuation factor and frequency are known, the low frequency oscillating discrete sampled signal model is written as:
Figure BDA0003024392830000091
in the formula, pi＝Aicosθi，qi＝AisinθiThe matrix expression of the low-frequency oscillation discrete sampling signal model is as follows:
x＝pC-bS；
wherein p ═ p1,p2,…,pQ]，q＝[q1,q2,…,qQ]，
Figure BDA0003024392830000092
Figure BDA0003024392830000093
Similarly, an error function is defined:
Figure BDA0003024392830000094
wherein x (n) is the actual sample;
Figure BDA0003024392830000095
is the output of the neural network. The performance index is defined as:
Figure BDA0003024392830000096
as the weight vector of the Adaline neural network, C, S is used as the input vector of the neural network. According to the training principle of the steepest descent method, the weight vectors p and q are adjusted as follows:
Figure BDA0003024392830000097
Figure BDA0003024392830000098
when the training of the Adaline neural network is finished, the amplitude and the phase of the oscillation mode are solved by the obtained weight vector and the following formula:
Figure BDA0003024392830000101
another objective of the present invention is to provide an electric power system oscillation parameter identification system using the electric power system oscillation parameter identification method, the electric power system oscillation parameter identification system comprising:
the oscillation signal separation module is used for decomposing an original signal into a plurality of filtering signals of different frequency bands by using a plurality of band-pass filters (BPFs) so as to realize the separation of oscillation signals of different frequency bands;
the oscillation mode signal extraction module is used for extracting oscillation mode signals from the BPF filtered signals by using an IVMD algorithm;
and the oscillation mode signal identification module is used for identifying the oscillation mode signal provided by the IVMD through an HTLS-Adaline algorithm to acquire information of each oscillation mode.
Another object of the present invention is to provide an information data processing terminal, characterized in that the information data processing terminal includes a memory and a processor, the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the power system oscillation parameter identification method.
Another object of the present invention is to provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method for identifying oscillation parameters of a power system.
By combining all the technical schemes, the invention has the advantages and positive effects that: aiming at signals containing a pan-band oscillation Mode, the method for identifying oscillation parameters of the power system provided by the invention firstly realizes the separation of signals in different frequency bands through a band-pass filter, then extracts signals of each oscillation Mode by utilizing an Improved Variable Mode Decomposition (IVMD) method with high noise robustness, and then estimates the frequency, attenuation factor, amplitude and phase of low-frequency oscillation by utilizing the combination of Hankel Total Least Square (HTLS) and an adaptive neural network algorithm (adaptive ANN). The introduction of the Adaline neural network solves the difficulty that the mode amplitude and the phase are not easy to determine after IVMD processing, and improves the detection precision. Simulation and practical example analysis show that the method can effectively distinguish and extract different types of oscillation modes in the signal and accurately identify the information of each mode. The IVMD-HTLS-Adaline method can effectively carry out mode identification on signals which are subjected to violent oscillation or noise-like signals containing potential oscillation.
The invention provides a method for identifying the pan-band oscillation of a complex power system based on an Improved VMD (Improved VMD, IVMD) method. Firstly, different types of oscillations are separated through a band-pass filter; secondly, extracting oscillation mode signals in the band-pass filtering signals by using the IVMD; and finally, respectively estimating the frequency, the attenuation factor, the amplitude and the phase of the low-frequency oscillation by combining an HTLS (Hankel total least squares) algorithm and an adaptive artificial Neural Network (Adaline ANN), thereby realizing the uniform identification of the pan-band oscillation. The performance of the algorithm is verified through a simulation example and an actual measurement example.
The invention provides a VMD-based power system universal frequency band oscillation signal extraction and mode identification method, which can effectively perform oscillation distinguishing, mode extraction and parameter identification on multi-type and universal frequency band oscillations such as low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation and the like from power system operation data. The identification result plays an important role in the aspects of system dynamic stability analysis and control, oscillation source identification and positioning and the like, and is beneficial to early warning and timely inhibiting system oscillation.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a method for identifying oscillation parameters of an electric power system according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a method for identifying oscillation parameters of an electric power system according to an embodiment of the present invention.
FIG. 3 is a block diagram of an embodiment of an oscillation parameter identification system of an electrical power system;
in the figure: 1. an oscillation signal separation module; 2. an oscillation mode signal extraction module; 3. and an oscillation mode signal identification module.
FIG. 4 is a flow chart of parameter optimization based on genetic variation particle swarm optimization provided by the embodiment of the present invention.
Fig. 5 is a schematic structural diagram of an adaptive linear neural network according to an embodiment of the present invention.
FIG. 6 shows a test signal y according to an embodiment of the present inventiontestSchematic time domain form.
Fig. 7 is a schematic diagram of oscillation mode signals of each frequency band according to an embodiment of the present invention.
Fig. 8 is a diagram illustrating raw oscillation signal data for identification according to an embodiment of the present invention.
Fig. 9 is a schematic diagram of extracting a mode signal from raw oscillation signal data according to an embodiment of the present invention.
Fig. 10 is a schematic diagram of signals obtained by linearly superimposing extracted modal signals and reconstructing the signals according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
In view of the problems in the prior art, the present invention provides a method, a system, a medium and a device for identifying oscillation parameters of a power system, which are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the method for identifying oscillation parameters of an electric power system according to an embodiment of the present invention includes the following steps:
s101, separating signals of different frequency bands through a band-pass filter;
s102, extracting oscillation mode signals in each band-pass filtering signal by using an Improved Variational Mode Decomposition (IVMD) method;
s103, combining the HTLS algorithm and the adaptive artificial neural network algorithm, respectively estimating the frequency, attenuation factor, amplitude and phase of the low-frequency oscillation, and realizing the uniform identification of the pan-band oscillation.
A schematic diagram of a method for identifying oscillation parameters of a power system according to an embodiment of the present invention is shown in fig. 2.
As shown in fig. 3, the system for identifying oscillation parameters of an electric power system according to an embodiment of the present invention includes:
the oscillation signal separation module 1 is configured to decompose an original signal into a plurality of filtering signals of different frequency bands by using a plurality of band-pass filters BPFs, so as to separate oscillation signals of different frequency bands;
the oscillation mode signal extraction module 2 is used for extracting an oscillation mode signal from each BPF filtered signal by using an IVMD algorithm;
and the oscillation mode signal identification module 3 is used for identifying the oscillation mode signal provided by the IVMD through an HTLS-Adaline algorithm to acquire information of each oscillation mode.
The technical solution of the present invention will be further described with reference to the following examples.
1. Summary of the invention
The oscillation problem becomes one of the important problems faced by the modern power grid, and various types of oscillation in the power system can occur simultaneously and the frequency band span is very large. Aiming at signals containing a pan-band oscillation Mode, the invention firstly realizes the separation of signals in different frequency bands through a band-pass filter, then extracts each oscillation Mode signal by utilizing an Improved Variant Mode Decomposition (IVMD) method with high noise robustness, and then estimates the frequency, attenuation factor, amplitude and phase of low-frequency oscillation by utilizing the combination of Hand Total Least Square (HTLS) and an adaptive neural network algorithm (Adaline ANN). The introduction of the Adaline neural network solves the difficulty that the mode amplitude and the phase are not easy to determine after IVMD processing, and improves the detection precision. Simulation and practical example analysis show that the method can effectively distinguish and extract different types of oscillation modes in the signal, and accurately identify the information of each mode. The IVMD-HTLS-Adaline method can effectively carry out mode identification no matter a signal which is subjected to severe oscillation or a noise-like signal containing potential oscillation.
The invention provides a method for identifying the pan-band oscillation of a complex power system based on an Improved VMD (Improved VMD, IVMD) method. Firstly, different types of oscillations are separated through a band-pass filter; secondly, extracting oscillation mode signals in the band-pass filtering signals by using the IVMD; and finally, respectively estimating the frequency, the attenuation factor, the amplitude and the phase of the low-frequency oscillation by combining an HTLS (Hankel total least squares) algorithm and an adaptive artificial Neural Network (Adaline ANN), thereby realizing the uniform identification of the pan-band oscillation. The performance of the algorithm is verified through a simulation example and an actual measurement example.
2. Universal band oscillation mode identification framework
2.1 broad band Oscillating Signal
The oscillation may include oscillation mode signals of different frequency bands andmeasuring the noise signal generated or present in the system itself. Raw measurement signal y with overtone oscillation0(t) can be expressed as:
Figure BDA0003024392830000141
in the formula, yNoise(t) is a noise signal; y isn(t) (N is 1,2, …, N) is a modal signal of N different frequency bands, and:
Figure BDA0003024392830000142
in the formula, An(t) is the amplitude of the nth mode of oscillation at time t, θnIs the initial phase; dnAnd fnRespectively damping ratio and oscillation frequency of the oscillation.
2.2 Modal recognition framework
The framework of the pan-band oscillation mode recognition is shown in fig. 2. The identification process is divided into the following three steps.
Step-1, decomposing an original signal into a plurality of filtering signals of different frequency bands by using a plurality of Band Pass Filters (BPF), thereby realizing the separation of oscillation signals of different frequency bands;
step-2, extracting oscillation mode signals from the BPF filtered signals by using an IVMD algorithm;
and Step-3, identifying the oscillation mode signals provided by the IVMD through an HTLS-Adaline algorithm, and acquiring information of each oscillation mode. The modal parameters can be obtained by filtering the original signals through a band-pass filter, extracting IVMD modal signals and identifying HTLS-Adaline.
3. Universal band oscillation mode identification based on VMD-HTLS-Adaline
3.1 bandpass Filter design
At present, the oscillations causing accidents in electric power systems are mainly classified into three types, namely: low frequency oscillation, subsynchronous oscillation, and supersynchronous oscillation. Wherein the low frequency oscillations are further divided into local and interval modes. The frequency of each type of oscillation is in a different range. Therefore, the present invention designs the BPFs to 4. The selection of the proper signal length can improve the accuracy of pattern recognition and ensure the rapidity of recognition. Since the oscillation frequencies are different, it is obviously not suitable to use data with the same duration for identification. The BPF filtering bandwidth and the identification sampling time length corresponding to various types of oscillation are shown in the table 1.
TABLE 1 BPF Filter Bandwidth and identification sampling duration for various oscillations
Figure BDA0003024392830000143
Figure BDA0003024392830000151
Through analysis of a modal identification algorithm, if the oscillation frequency and the oscillation amplitude are required to be accurately obtained, the sampling time of an identification signal is required to be more than 1 oscillation period; in order to obtain accurate damping ratio information, the sampling time is required to be longer than 2 oscillation periods. Therefore, aiming at low-frequency oscillation, in order to ensure the rapidity of identification, the invention takes the lowest band-pass frequency of the band-pass filter of 0.2Hz as the reference, and the sampling duration is selected to be 2 times of the oscillation period, namely 10 s; considering the subsynchronous oscillation, the comprehensive precision and the identification speed, and taking the lowest oscillation frequency of 10Hz as a reference, wherein the sampling duration is 4 times of the oscillation period, namely 0.4 s; the mechanism of the super-synchronous oscillation determines that the super-synchronous oscillation and the sub-synchronous oscillation occur in pairs and have complementary frequencies, so that the sampling time of the super-synchronous oscillation is consistent with that of the sub-synchronous oscillation and is also 0.4 s.
3.2 Modal Signal extraction based on improved VMD Algorithm
3.2.1VMD Algorithm
The VMD algorithm is a novel self-adaptive signal decomposition new method proposed by Dragomiretsky et al in 2014, and a target mode is solved through an inherent mode function. In view of the accuracy and noise robustness of the VMD algorithm, the method adopts the VMD method to extract and separate modal signals. VMD algorithm correspondenceThe variation problem of (1) is to seek the K IMF components u whose sum of the estimated bandwidths is minimumk(t) of (d). Transforming the variation problem to obtain an augmented Lagrange equation, and solving the equation by adopting an alternating Direction multiplier (ADMM) of multiplications to obtain a mode function
Figure BDA0003024392830000152
Solving the value problem:
Figure BDA0003024392830000153
similarly, the solution to the problem of the center frequency of the modal component is:
Figure BDA0003024392830000154
wherein, { u [ [ u ] ]k}＝{u1,u2,…,uK}，{ωk}＝{ω1,ω2,…,ωKThe components and their center frequencies.
The algorithm flow of the VMD is as follows:
step 1 initialization
Figure BDA0003024392830000161
And n is 0;
step 2n ← n +1, and updates u according to expressions (1) and (2)kAnd ωk；
Step 3, updating lambda:
Figure BDA0003024392830000162
step 4 repeat steps 2 and 3 until the iteration stop condition is met, i.e.
Figure BDA0003024392830000163
Ending circulation, outputting the result to obtain K modal components and center frequencies thereofAnd (4) rate.
3.2.2 PSO Algorithm optimized VMD penalty factor determination
The penalty factor alpha in the VMD algorithm has a large influence on the decomposition result, and researches show that: the smaller the penalty parameter α is, the larger the bandwidth of each obtained imf (intrinsic Mode function) component is, and conversely, the larger α is, the smaller the bandwidth of each component is. Therefore, when the VMD is used to decompose the oscillation signal of the power system, it is very important to select an appropriate penalty factor parameter α. The method adopts a genetic variation particle swarm algorithm to optimize the punishment parameters and obtain the optimal alpha value.
The particle swarm optimization is a global optimization algorithm proposed by Eberh and Kennedy et al in 1995, the method is a group intelligent optimization algorithm, the method has the advantages of fewer parameters and easiness in adjustment, meanwhile, local optimization is easy to fall into, and a global optimal approximate solution cannot be obtained.
Defining a genetic variation particle swarm algorithm: in a search space of D dimension, the marker population is X, and the species X is composed of m particles, i.e., X ═ X1,x2,…,xm]The position of each particle in the search space can be represented by a D-dimensional vector, that is:
xi＝[xi1,xi2,…,xiD]d is the number of parameters to be optimized; moving speed v of ith particlei＝[vi1,vi2,…,viD]Local extremum p of the particlei＝[pi1,pi2,…,piD]Global extreme G of the generation population1＝[g1,g2,…,gDSub-global optimum G2＝[g′1,g′2,…,g′D′]The maximum optimal retention algebra of an individual is maxAge, and the mutation probability is q. In order to prevent the particles from falling into local optimum, the individually optimum keeping algebra of the particles in the iteration process needs to be recorded, when the individually optimum keeping algebra does not reach maxAge, each particle updates the position and the speed of the next generation through an individual local extreme value and a global extreme value, and a formula is updatedThe following were used:
Figure BDA0003024392830000171
Figure BDA0003024392830000172
where ω is the inertial weight and η is [0, 1 ]]Random number in between, c1And c2For learning factors, respectively representing local search capability and global search capability, n represents iteration number, v represents iteration numberi，pi，G1，xiAre all D-dimensional vectors. The determination of the inertia weight omega of the current iteration times adopts a linear decreasing weight method proposed by Shi, and the formula is as follows:
ω＝ωmax-(ωmax-ωmin)n/nmax (16)
in the formula, ωmaxAnd ωminRespectively, the maximum and minimum inertia weight, n is the current iteration number, nmaxIs a defined maximum number of iterations. When the individual optimum kept algebra to reach maxAge, the position and speed of the particle are updated by genetic variation operation to make the particle jump out of local optimum.
Selecting a fitness function of the genetic variation particle swarm algorithm: during parameter optimization, the evaluation standard of the decomposition effect of the VMD method adopts the envelope entropy E proposed by Tang Guiji et alpNotion, the envelope entropy of a time signal x (j) of length N is defined as:
Figure BDA0003024392830000173
where a (j) is x (j) the envelope signal demodulated by hillbert, and j is 1,2, …, N. p is a radical ofjIs the result after a (j) normalization, normalization not only avoids the influence of different envelope amplitudes of IMF components, but also reduces the interference of weak noise, EpIs obtained according to an information entropy calculation rule and is obtained according to the method EpMeasure score of VMDAnd (5) effect solving.
After the BPF-filtered oscillation signals are decomposed by a VMD method, if more noise is contained in the obtained components, the sparsity of the component signals is weaker, and the envelope entropy is larger; on the contrary, if regular oscillation signals appear in the components, the signals will present strong sparsity, and the calculated envelope entropy is small. Therefore, under the influence of the parameter alpha, the minimum envelope entropy E of the K components is selectedpAs local minimum entropy minEpThe component corresponding to the minimum entropy value contains abundant characteristic information. Using local minimum entropy as a part of fitness function in the whole search process, and searching a parameter alpha corresponding to the global optimum component0. Through the analysis of the parameter alpha, the number of iterations of the VMD is reduced by properly selecting alpha, namely the decomposition efficiency of the VMD method is higher. Therefore, the invention needs to achieve higher decomposition efficiency as much as possible when the decomposition effect is optimal, and the invention is in the minEpOn the basis, time (iteration times) is added, and the following fitness function is constructed:
minF＝minEp+β·time
where β is a quantization factor of the fitness function. In the present invention, β is 1/1000. The parameter optimization process based on the genetic variation particle swarm optimization is shown in FIG. 4.
The VMD extracted oscillation mode u can be obtained by providing signal decomposition for BPF through the improved VMDn(t) of (d). For modal signal u when BPF and VMD accuracy is high enoughn(t) having:
un(t)≈yn(t)
therefore, all oscillation mode signals can be separated from the original signals through the BPF and the IVMD.
3.3 solving for frequency and attenuation factor by IVMD-HTLS Algorithm
Obtaining each modal signal u obtained by IVMD decompositionnAfter (t), the mode parameters such as oscillation frequency, attenuation factor and the like can be identified through an HTLS algorithm. The HTLS algorithm belongs to a subspace rotation invariant method, and is high in operation efficiency and strong in noise resistance. The main idea is to construct 1 Hankel matrix by using sampling signals and to carry out van der waals on the matrixAnd (3) Mongolian decomposition, namely constructing an equivalent relation by using the translation invariant characteristic of the Van der Mongolian matrix, solving the characteristic value of the oscillation mode, and finally obtaining the frequency, the attenuation factor, the amplitude and the phase of each oscillation mode. The main idea of IVMD-HTLS is as follows: decomposing the BPF filtered sequence by using IVMD; then, for each decomposed component, the oscillation frequency and the attenuation factor are calculated by using an HTLS algorithm. Since the FOMC-HTLS algorithm cannot give the amplitude and the phase of the original signal y (n), the information of each mode is incomplete, and the reconstruction of the signal and the quantitative evaluation of the algorithm are not facilitated. Therefore, the invention introduces Adaline neural network to solve the oscillation mode information (amplitude and phase).
3.4Adaline neural network solving for amplitude and phase
3.4.1 adaptive Linear neural network
An adaptive linear (Adaline) neural network is a neuron model originally proposed by Widrow and Hoff, is widely applied to the fields of signal processing and the like, and the structural principle of the neural network is shown in fig. 5.
In FIG. 5, x1k,x2k,…,xnkN input signals at time k for the adaptive linear neural network. The input signal is represented in vector form as Xik＝[x1k,x2k,…,xnk]TOften referred to as the input mode vector of the Adaline neural network. The weight vector corresponding to each group of input signals is Wik＝[w1k,w2k,…,wnk]T. The Adaline neural network output is:
Figure BDA0003024392830000191
let y (k) be the ideal response signal, defining the error function as:
Figure BDA0003024392830000192
the Adaline neural network works as follows: the ideal response signal y (k) is connected with the output of the neural networkOutput signal
Figure BDA0003024392830000193
Comparing to obtain difference e (k), transmitting e (k) to learning rule, adjusting weight vector according to learning rule, and making
Figure BDA0003024392830000194
In agreement with y (k).
The learning rule of the Adaline neural network is a Widrow-Hoff rule, namely a least mean square deviation algorithm (LMS). The rule weight vector adjustment expression is:
Wi(k+1)＝Wi(k+1)+ηe(k)Xi(k+1)
in the formula, η is the learning rate of the Adaline neural network, and η belongs to (0,1), and the value directly influences the weight vector regulation precision and the convergence speed.
3.4.2Adaline neural network solution for oscillation mode
The detailed steps for the Adaline neural network to solve for amplitude and phase are as follows. The known low-frequency oscillation discrete sampling signal model is:
Figure BDA0003024392830000195
when the attenuation factor and frequency are known, equation (16) can be written as:
Figure BDA0003024392830000201
in the formula, pi＝Aicosθi，qi＝AisinθiThe matrix expression of equation (17) is:
x＝pC-bS
wherein p ═ p1,p2,…,pQ]，q＝[q1,q2,…,qQ]，
Figure BDA0003024392830000202
Figure BDA0003024392830000203
Similarly, an error function is defined:
Figure BDA0003024392830000204
wherein x (n) is the actual sample;
Figure BDA0003024392830000205
is the output of the neural network. The performance index is defined as:
Figure BDA0003024392830000206
since the attenuation factor and the frequency are known, p and q are unknowns in equation (18) and serve as weight vectors of the Adaline neural network, and C, S serve as input vectors of the neural network. According to the training principle of the steepest descent method, the weight vectors p and q are adjusted as follows:
Figure BDA0003024392830000207
Figure BDA0003024392830000208
when the training of the Adaline neural network is finished, the amplitude and the phase of the oscillation mode are solved by the obtained weight vector and the formula (23).
Figure BDA0003024392830000211
4. Simulation and actual example analysis
4.1 simulation analysis
To verify the effectiveness of the method of the invention, the oscillating signal y is constructed by simulationtest：
ytest＝yLFO1+yLFO2+ySSO1+ySSO2+ySurSO+yNoise；
In the formula, yLFO1A low-frequency oscillation interval mode is adopted; y isLFO2A low-frequency oscillation local mode; y isSSO1Is sub-synchronous oscillation and is in accordance with the super-synchronous oscillation signal ySurSOPairing; y isSSO2Is another independent subsynchronous oscillation; y isNoiseIs white noise. To be compatible with the real situation, the test signal has satisfied the following conditions and assumptions: (1) all oscillation frequencies are randomly selected in the oscillation type frequency band to which the oscillation frequencies belong; (2) the low-frequency oscillation is a main oscillation mode, the amplitude is higher than the subsynchronous oscillation, and the amplitude of a local mode of the low-frequency oscillation is higher than that of an interval mode; (3) in the subsynchronous and supersynchronous oscillation occurring in pairs, the amplitude of the subsynchronous oscillation mode is slightly higher than that of the supersynchronous mode. Based on the above assumptions, the specific parameters of each mode of the finally selected test signal are:
Figure BDA0003024392830000212
i.e. the test signal ytestWhich contains 5 oscillation signals of different frequencies and a white noise signal with an amplitude of 0.16. Test signal ytestThe oscillation mode frequency range covers 0.63-93.42 Hz. Each test signal is oscillating with a constant amplitude. Test signal ytestThe time domain form is shown in fig. 6. By using the method of the invention, the constructed test signal ytestAfter band-pass filtering and IVMD extraction, the oscillation mode signals of each frequency band are shown in fig. 7. As can be seen from fig. 7, the method provided by the present invention can accurately distinguish the modes of different frequency bands, and effectively extract all the corresponding mode signals.
Modal signals are extracted from each IVMD obtained in the figure 7, and modal identification is carried out through an HTLS-Adaline algorithm, so that parameter information of corresponding modes can be obtained. The identification results of the method of the present invention were compared with those of other methods, as shown in table 2. The selected comparison method is as follows: (1) the classic EMD-based HTLS algorithm (EMD-HTLS); (2) a VMD-based HTLS algorithm (VMD-HTLS); (3) the method is based on IVMD-HTLS and Adaline algorithms (IVMD-HTLS-Adaline).
TABLE 2 comparison of the identification results of the present invention method with those of other methods
Figure BDA0003024392830000221
From the comparison in table 2, it can be seen that the method of the present invention can effectively identify all oscillation modes when the test signal oscillates across multiple frequency bands. The maximum error of the oscillation frequency identification result is 1.55%, and the average error is 0.67%; the maximum error of the oscillation amplitude identification result is 9.09%, and the average error is 2.54%. In contrast, the EMD-HTLS method has the worst frequency identification and amplitude identification results and the highest average error. The VMD-HTLS method dominates the oscillation mode (y) with the highest amplitudeLFO2) The method has good identification effect, the frequency and amplitude identification result is close to that of the method, but the phase identification result is not as good as that of the method; for the secondary mode yLFO1Compared with ySSO1, the VMD-HTLS frequency identification effect is close to that of the method, but the amplitude and phase identification results are far worse than that of the method provided by the invention. In particular for low amplitude, high frequency modes ySSO2The EEMD-HTLS and VMD-HTLS have poor recognition results. For a supersynchronous oscillation mode y with a frequency of 93.42SurSOThe EEMD-HTLS method is not recognized, and the recognition result of the VMD-HTLS method is inconsistent with the actual result. Therefore, the method provided by the invention has a remarkable advantage in the identification of the multi-modal concurrent pan-band oscillation compared with other methods.
4.2 actual example data
In order to prove the effectiveness of the method, the actual oscillation data of the power grid in Hunan is selected, and the method provided by the invention is adopted to carry out oscillation mode identification analysis. The oscillation events are as follows: 24 days 6 months 2018, a certain thermal power plant in the Hunan power grid generates low-frequency oscillation. The system starts to generate low-frequency oscillation at 120s, the oscillation quickly diverges after 30s, the system starts to give an alarm, and the oscillation is maintained for about 180 s. And (5) in 60-120 s before the oscillation occurs, the system has noise-like oscillation, and the system does not give an alarm at the moment. The oscillation signal and noise-like signal intervals are indicated in the figure. The sampling frequency of the system is 100 Hz. And selecting the oscillation signals of 165-195 s, and performing modal identification on the oscillation signals by adopting the method. The original signal for identification is shown in fig. 8.
As can be seen from fig. 8, the amplitude of the oscillation signal is high, the maximum frequency fluctuation exceeds 100MW, and is higher than 30% of the system output power, and the signal is affected by a certain degree of noise.
For the raw oscillation signal data shown in fig. 8, the modal signal extracted by the algorithm of the present invention is shown in fig. 9. The method of the invention separates 3 modes from the oscillation signal data, namely the main mode of system oscillation. Of the 3 modes, the 1 st mode has the highest amplitude and is the oscillation dominant mode. The extracted modal signals are linearly superimposed, and the reconstructed signals are shown in fig. 10.
Comparing fig. 10 with fig. 8, it is easy to find that the reconstructed signal substantially coincides with the original signal waveform. That is, the separated 3 mode signals cover the main oscillation of the system, and the method of the invention can effectively extract all the main oscillation mode signals in the signals.
The information of the main oscillation mode can be obtained by adopting the Prony algorithm to carry out parameter identification on the mode signals of the graph 9. The results of the comparison of the method of the present invention with MF-Prony and EMD-Prony parameter identification are shown in Table 3. As can be seen from the comparison, the mode with the highest amplitude can be identified for the oscillation signal by the method, MF-Prony or EMD-Prony. The MF-Prony frequency identification error is the highest in the three methods, and the result of the method is similar to that of the EMD-Prony method. During the violent oscillation period of the system, the frequency of the dominant oscillation mode is about 1.48Hz, and the frequency of the secondary dominant mode is about 2.02 Hz.
TABLE 3 comparison of the identification results of the MF-Prony and EMD-Prony parameters in the method of the present invention
Figure BDA0003024392830000231
5. Conclusion
The invention provides a VMD-based power system universal frequency band oscillation signal extraction and mode identification method, which can effectively perform oscillation distinguishing, mode extraction and parameter identification on multi-type and universal frequency band oscillations such as low-frequency oscillation, subsynchronous oscillation, supersynchronous oscillation and the like from power system operation data. The identification result plays an important role in the aspects of system dynamic stability analysis and control, oscillation source identification and positioning and the like, and is beneficial to early warning and timely inhibiting system oscillation.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any modification, equivalent replacement, and improvement made by those skilled in the art within the technical scope of the present invention disclosed in the present invention should be covered within the scope of the present invention.

Claims (10)

1. A method for identifying oscillation parameters of a power system is characterized by comprising the following steps:
separating signals of different frequency bands through a band-pass filter;
extracting oscillation mode signals in each band-pass filtering signal by using an improved variational mode decomposition method;
and estimating the frequency, attenuation factor, amplitude and phase of the low-frequency oscillation to realize the uniform identification of the pan-band oscillation.
2. The method for identifying oscillation parameters of an electric power system as claimed in claim 1, wherein the method for identifying oscillation parameters of an electric power system comprises the following steps:
using a plurality of band-pass filters (BPFs) to decompose an original signal into a plurality of filtering signals of different frequency bands, so as to realize the separation of oscillation signals of different frequency bands;
extracting oscillation mode signals from the BPF filtered signals by using an IVMD algorithm;
and step three, identifying oscillation mode signals provided by the IVMD through an HTLS-Adaline algorithm, and acquiring information of each oscillation mode.
3. A method for identifying oscillation parameters of a power system as defined in claim 2, wherein in the first step, the oscillation includes oscillation mode signals of a plurality of different frequency bands and noise signals generated by measurement or existing in the system; wherein the raw measurement signal y containing the pan-band oscillation0(t) is expressed as:
Figure FDA0003024392820000011
in the formula, yNoise(t) is a noise signal; y isn(t) (N is 1,2, …, N) is a modal signal of N different frequency bands, and:
Figure FDA0003024392820000012
in the formula, An(t) is the amplitude of the nth mode of oscillation at time t, θnIs the initial phase; dnAnd fnRespectively damping ratio and oscillation frequency of the oscillation.
4. The method according to claim 2, wherein in the first step, the band pass filters are designed to be 4, and the identifying the sampling time duration and the BPF filtering bandwidth corresponding to each type of oscillation comprises:
firstly, a band-pass filter BPF-1 is adopted, the oscillation type is an LFO interval mode, the sampling time length is 10s, and the oscillation frequency band is 0.2-1 Hz;
secondly, a band-pass filter BPF-2 is adopted, the oscillation type is an LFO local mode, the sampling time is 2s, and the oscillation frequency band is 1-5 Hz;
thirdly, a band-pass filter BPF-3 is adopted, the oscillation type is SSO, the sampling time is 0.4s, and the oscillation frequency band is 10-50 Hz;
fourthly, a band-pass filter BPF-4, wherein the oscillation type is SurSO, the sampling time is 0.4s, and the oscillation frequency band is 70-110 Hz;
taking the lowest band-pass frequency of the band-pass filter of 0.2Hz as a reference, and selecting the sampling duration to be 2 times of the oscillation period, namely 10 s; for subsynchronous oscillation, the comprehensive precision and the identification speed are considered, the lowest oscillation frequency of 10Hz is taken as the reference, and the sampling duration is 4 times of the oscillation period, namely 0.4 s; the mechanism of the super-synchronous oscillation determines that the super-synchronous oscillation and the sub-synchronous oscillation appear in pairs and have complementary frequencies, so that the sampling time of the super-synchronous oscillation is consistent with that of the sub-synchronous oscillation and is 0.4 s.
5. The method for identifying oscillation parameters of an electric power system as claimed in claim 2, wherein in the second step, the mode signal extraction based on the improved VMD algorithm comprises:
the variational problem corresponding to the VMD algorithm is to seek the K IMF components u with the minimum sum of the estimation bandwidthsk(t); transforming the variation problem to obtain an augmented Lagrange equation, and solving the equation by adopting a multiplier Alternating Direction Multiplier Method (ADMM) to obtain a modal function
Figure FDA0003024392820000021
Solving the value problem:
Figure FDA0003024392820000022
similarly, the solution to the problem of the center frequency of the modal component is:
Figure FDA0003024392820000023
wherein, { u [ [ u ] ]k}＝{u1,u2,…,uK}，{ωk}＝{ω1,ω2,…,ωK-each component and its center frequency;
the algorithm flow of the VMD is as follows:
(1) initialization
Figure FDA0003024392820000031
And n is 0;
(2) n ← n +1, and u is updated from expressions (1) and (2)kAnd ωk；
(3) Updating lambda:
Figure FDA0003024392820000032
(4) repeating steps (2) and (3) until an iteration stop condition is satisfied, i.e.
Figure FDA0003024392820000033
And ending circulation, and outputting a result to obtain K modal components and the center frequency thereof.
6. The method for identifying oscillation parameters of an electric power system as claimed in claim 2, wherein in the second step, the mode signal extraction based on the improved VMD algorithm further comprises: determining a VMD penalty factor based on PSO algorithm optimization;
the penalty factor alpha in the VMD algorithm has a large influence on the decomposition result, and researches show that: the smaller the penalty parameter alpha is, the larger the bandwidth of each obtained IMF component is, and on the contrary, the larger the alpha is, the smaller the bandwidth of each component is; therefore, when the VMD is used for decomposing the oscillation signals of the power system, it is very important to select a proper penalty factor parameter alpha; optimizing the punishment parameters by adopting a genetic variation particle swarm algorithm to obtain an optimal alpha value;
introducing the idea of genetic algorithm variation into the particle swarm algorithm, and constructing the genetic variation particle swarm algorithm;
defining a genetic variation particle swarm algorithm: in a search space of D dimension, the labeled population is X, and the population X is composed of m particles, i.e., X ═ X1,x2,…,xm]The position of each particle in the search space can be represented by a D-dimensional vector, that is:
xi＝[xi1,xi2,…,xiD]d is the number of parameters to be optimized; moving speed v of ith particlei＝[vi1,vi2,…,viD]Local extremum p of the particlei＝[pi1,pi2,…,piD]Global extreme G of the generation population1＝[g1,g2,…,gDSub-global optimum G2＝[g′1,g′2,…,g′D]The maximum optimal retention algebra of an individual is maxAge, and the mutation probability is q; in order to prevent the particles from falling into local optimality, the individually optimal kept algebra of the particles in the iterative process needs to be recorded, when the individually optimal kept algebra does not reach maxAge, each particle updates the position and the speed of the next generation through an individual local extremum and a global extremum, and the updating formula is as follows:
Figure FDA0003024392820000034
Figure FDA0003024392820000041
where ω is the inertial weight and η is [0, 1 ]]Random number in between, c1And c2For learning factors, the local search capability and the global search capability are respectively represented, n represents the iteration number, v represents the iteration numberi，pi，G1，xiAre all D-dimensional vectors; the determination of the inertia weight omega of the current iteration times adopts a linear decreasing weight method proposed by Shi, and the formula is as follows:
ω＝ωmax-(ωmax-ωmin)n/nmax
in the formula, ωmaxAnd ωminRespectively, the maximum and minimum inertia weight, n is the current iteration number, nmaxIs a defined maximum number of iterations; when the optimal kept algebra of the individual reaches maxAge, updating the position and the speed of the particle by adopting genetic variation operation to enable the particle to jump out of the local optimal;
selecting a fitness function of the genetic variation particle swarm algorithm: in parameter optimization, the evaluation standard of the decomposition effect of the VMD method adopts the envelope entropy E proposed by Tang Guiji et alpNotion, the envelope entropy of a time signal x (j) of length N is defined as:
Figure FDA0003024392820000042
where a (j) is x (j) the envelope signal demodulated by hillbert, j is 1,2, …, N; p is a radical ofjIs the result after normalization of a (j), EpIs obtained according to an information entropy calculation rule according to EpMeasuring the decomposition effect of the VMD;
after the BPF-filtered oscillation signals are decomposed by a VMD method, if more noise is contained in the obtained components, the sparsity of the component signals is weaker, and the envelope entropy is larger; on the contrary, if the components have regular oscillation signals, the signals have strong sparsity, and the calculated envelope entropy is small; therefore, under the influence of the parameter alpha, the minimum envelope entropy E of the K components is selectedpAs local minimum entropy minEpThe component corresponding to the minimum entropy value contains abundant characteristic information; using local minimum entropy as a part of fitness function in the whole search process, and searching a parameter alpha corresponding to the global optimum component0(ii) a Through the analysis of the parameter alpha, the number of iterations of the VMD is reduced by properly selecting the alpha, namely the decomposition efficiency of the VMD method is higher; at minEpOn the basis, adding iteration time to construct a fitness function as follows:
minF＝minEp+β·time；
wherein, beta is a quantization factor of the fitness function, and beta is selected as 1/1000;
the VMD extracted oscillation mode u can be obtained by providing signal decomposition for BPF through the improved VMDn(t); for modal signal u when BPF and VMD accuracy is high enoughn(t) having:
un(t)≈yn(t)；
therefore, all oscillation mode signals can be separated from the original signals through the BPF and the IVMD.
7. The method according to claim 2, wherein the solving of the frequency and the attenuation factor by the IVMD-HTLS algorithm in step three comprises:
(1) adaline neural network solution for amplitude and phase
In adaptive linear neural networks, x1k,x2k,…,xnkN input signals at time k for the adaptive linear neural network; the vector form of the input signal is represented as Xik＝[x1k,x2k,…,xnk]TInput pattern vectors, often referred to as Adaline neural networks; the weight vector corresponding to each group of input signals is Wik＝[w1k,w2k,…,wnk]T(ii) a The Adaline neural network output is:
Figure FDA0003024392820000051
let y (k) be the ideal response signal, defining the error function as:
Figure FDA0003024392820000052
the Adaline neural network works as follows: the ideal response signal y (k) is compared with the output signal of the neural network
Figure FDA0003024392820000053
Comparing to obtain difference e (k), transmitting e (k) to learning rule, adjusting weight vector according to learning rule, and making
Figure FDA0003024392820000054
In agreement with y (k);
the learning rule of the Adaline neural network is a Widrow-Hoff rule, namely a least square mean square error algorithm LMS; the rule weight vector adjustment expression is:
Wi(k+1)＝Wi(k+1)+ηe(k)Xi(k+1)；
in the formula, eta is the learning rate of the Adaline neural network, and eta belongs to (0,1), and the value of eta directly influences the adjustment precision and the convergence speed of the weight vector;
(2) adaline neural network solution for oscillation modes
The detailed steps of the Adaline neural network for solving the amplitude and the phase are as follows:
the known low-frequency oscillation discrete sampling signal model is:
Figure FDA0003024392820000061
when the attenuation factor and frequency are known, the low frequency oscillating discrete sampled signal model is written as:
Figure FDA0003024392820000062
in the formula, pi＝Aicosθi，qi＝AisinθiThe matrix expression of the low-frequency oscillation discrete sampling signal model is as follows:
x＝pC-bS；
wherein p ═ p1,p2,…,pQ]，q＝[q1,q2,…,qQ]，
Figure FDA0003024392820000063
Figure FDA0003024392820000064
Similarly, an error function is defined:
Figure FDA0003024392820000065
wherein x (n) is the actual sample;
Figure FDA0003024392820000066
is the output of the neural network; the performance index is defined as:
Figure FDA0003024392820000067
as the weight vector of Adaline neural network, C, S as the input vector of neural network; according to the training principle of the steepest descent method, the weight vectors p and q are adjusted as follows:
Figure FDA0003024392820000071
Figure FDA0003024392820000072
when the training of the Adaline neural network is finished, the amplitude and the phase of the oscillation mode are solved by the obtained weight vector and the following formula:
Figure FDA0003024392820000073
8. an electric power system oscillation parameter identification system applying the electric power system oscillation parameter identification method according to any one of claims 1 to 7, the electric power system oscillation parameter identification system comprising:
the oscillation signal separation module is used for decomposing an original signal into a plurality of filtering signals of different frequency bands by using a plurality of band-pass filters (BPFs) so as to realize the separation of oscillation signals of different frequency bands;
the oscillation mode signal extraction module is used for extracting oscillation mode signals from the BPF filtered signals by using an IVMD algorithm;
and the oscillation mode signal identification module is used for identifying the oscillation mode signal provided by the IVMD through an HTLS-Adaline algorithm to acquire information of each oscillation mode.
9. An information data processing terminal, characterized in that the information data processing terminal comprises a memory and a processor, the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the power system oscillation parameter identification method according to any one of claims 1 to 7.
10. A computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform a power system oscillation parameter identification method as claimed in any one of claims 1 to 7.
CN202110411632.7A 2021-04-16 2021-04-16 Electric power system oscillation parameter identification method, system, terminal and storage medium Pending CN113127801A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110411632.7A CN113127801A (en) 2021-04-16 2021-04-16 Electric power system oscillation parameter identification method, system, terminal and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110411632.7A CN113127801A (en) 2021-04-16 2021-04-16 Electric power system oscillation parameter identification method, system, terminal and storage medium

Publications (1)

Publication Number Publication Date
CN113127801A true CN113127801A (en) 2021-07-16

Family

ID=76776824

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110411632.7A Pending CN113127801A (en) 2021-04-16 2021-04-16 Electric power system oscillation parameter identification method, system, terminal and storage medium

Country Status (1)

Country Link
CN (1) CN113127801A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113489026A (en) * 2021-07-26 2021-10-08 东南大学 Wide-frequency oscillation identification method and positioning method for wide-area power system
CN114552580A (en) * 2022-04-25 2022-05-27 东南大学溧阳研究院 Intelligent sub/super synchronous oscillation identification method based on synchronous phasor
CN115931358A (en) * 2023-02-24 2023-04-07 沈阳工业大学 Bearing fault acoustic emission signal diagnosis method with low signal-to-noise ratio
CN116720067A (en) * 2023-06-02 2023-09-08 广东省麦思科学仪器创新研究院 Mass spectrogram global peak information feature description method and device based on oscillation signals
CN117725394A (en) * 2024-02-18 2024-03-19 浙江浙能技术研究院有限公司 Wind power plant broadband oscillation identification method based on hierarchical embedded modal decomposition

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111781434A (en) * 2019-04-03 2020-10-16 国网湖南省电力有限公司 Universal frequency band oscillation unified identification method for power system

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111781434A (en) * 2019-04-03 2020-10-16 国网湖南省电力有限公司 Universal frequency band oscillation unified identification method for power system

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
刘建昌 等: "基于参数优化 VMD 和样本熵的滚动轴承故障诊断", 《自动化学报》 *
唐贵基 等: "参数优化变分模态分解方法在滚动轴承早期故障诊断中的应用", 《西安交通大学学报》 *
张虹 等: "含高渗透率水电机组的电力系统振荡参数辨识", 《电力系统及其自动化学报》 *
汤吉鸿 等: "基于变分模态分解的电力系统泛频带振荡辨识方法", 《电力系统保护与控制》 *
王臻 等: "考虑高斯有色噪声的FOMC-HTLS-Adaline算法在低频振荡模式辨识中的研究", 《电工技术学报》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113489026A (en) * 2021-07-26 2021-10-08 东南大学 Wide-frequency oscillation identification method and positioning method for wide-area power system
CN113489026B (en) * 2021-07-26 2023-08-08 东南大学 Wide-band oscillation identification method and positioning method for wide-area power system
CN114552580A (en) * 2022-04-25 2022-05-27 东南大学溧阳研究院 Intelligent sub/super synchronous oscillation identification method based on synchronous phasor
CN114552580B (en) * 2022-04-25 2022-07-26 东南大学溧阳研究院 Intelligent sub/super synchronous oscillation identification method based on synchronous phasor
CN115931358A (en) * 2023-02-24 2023-04-07 沈阳工业大学 Bearing fault acoustic emission signal diagnosis method with low signal-to-noise ratio
CN115931358B (en) * 2023-02-24 2023-09-12 沈阳工业大学 Bearing fault acoustic emission signal diagnosis method with low signal-to-noise ratio
CN116720067A (en) * 2023-06-02 2023-09-08 广东省麦思科学仪器创新研究院 Mass spectrogram global peak information feature description method and device based on oscillation signals
CN117725394A (en) * 2024-02-18 2024-03-19 浙江浙能技术研究院有限公司 Wind power plant broadband oscillation identification method based on hierarchical embedded modal decomposition
CN117725394B (en) * 2024-02-18 2024-05-07 浙江浙能技术研究院有限公司 Wind power plant broadband oscillation identification method based on hierarchical embedded modal decomposition

Similar Documents

Publication Publication Date Title
CN113127801A (en) Electric power system oscillation parameter identification method, system, terminal and storage medium
Ciancetta et al. A new convolutional neural network-based system for NILM applications
Liu et al. A novel three-step classification approach based on time-dependent spectral features for complex power quality disturbances
CN109948597B (en) High-voltage circuit breaker mechanical fault diagnosis method
CN110648088B (en) Electric energy quality disturbance source judgment method based on bird swarm algorithm and SVM
CN111027408A (en) Load identification method based on support vector machine and V-I curve characteristics
CN111368892B (en) Electric energy quality disturbance efficient identification method for generalized S transformation and SVM
CN109389325B (en) Method for evaluating state of electronic transformer of transformer substation based on wavelet neural network
CN116559598B (en) Smart distribution network fault positioning method and system
CN109359665A (en) A kind of family's electric load recognition methods and device based on support vector machines
CN112213687B (en) Gateway electric energy meter data anomaly detection method and system based on pseudo-anomaly point identification
CN112183400B (en) Novel latent fault feature extraction method and system for distribution transformer
CN111639583A (en) Method and system for identifying power quality disturbance of power grid
CN110244185A (en) A kind of multi-source harmonic contributions division methods, terminal device and storage medium
Ray et al. A comprehensive review on soft computing and signal processing techniques in feature extraction and classification of power quality problems
Porat et al. Adaptive detection of transient signals
CN115600088A (en) Distribution transformer fault diagnosis method based on vibration signals
CN105954695A (en) Synchronization-based homogeneous-sensor mutation parameter recognizing method and device
CN112636328B (en) Medium-voltage distribution network equipment load state identification method
CN110146922A (en) The double seismometer interference identification methods of high speed rail earthquake pre-warning system list
CN109635430A (en) Grid power transmission route transient signal monitoring method and system
CN116681186A (en) Power quality analysis method and device based on intelligent terminal
Haugdal et al. Mode shape estimation using complex principal component analysis and k-means clustering
Chiam et al. Noise level evaluation on power quality disturbances classification
Yang et al. Transient fault diagnosis of track circuit based on MFCC-DTW

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: 20210716

RJ01 Rejection of invention patent application after publication