CN111398679B - Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit) - Google Patents

Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit) Download PDF

Info

Publication number
CN111398679B
CN111398679B CN202010158721.0A CN202010158721A CN111398679B CN 111398679 B CN111398679 B CN 111398679B CN 202010158721 A CN202010158721 A CN 202010158721A CN 111398679 B CN111398679 B CN 111398679B
Authority
CN
China
Prior art keywords
data
amplitude
oscillation
inter
sequence
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.)
Active
Application number
CN202010158721.0A
Other languages
Chinese (zh)
Other versions
CN111398679A (en
Inventor
刘灏
亓源
毕天姝
熊文
王莉
危国恩
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.)
North China Electric Power University
Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
Original Assignee
North China Electric Power University
Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd
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 North China Electric Power University, Guangzhou Power Supply Bureau of Guangdong Power Grid Co Ltd filed Critical North China Electric Power University
Priority to CN202010158721.0A priority Critical patent/CN111398679B/en
Publication of CN111398679A publication Critical patent/CN111398679A/en
Application granted granted Critical
Publication of CN111398679B publication Critical patent/CN111398679B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R25/00—Arrangements for measuring phase angle between a voltage and a current or between voltages or currents
    • 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
    • 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
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a method for identifying and alarming subsynchronous oscillation based on PMU (phasor measurement unit), which is used for realizing the rapid identification of the subsynchronous oscillation based on a support vector machine algorithm, filtering inter-harmonic components caused by noise based on FFT (fast Fourier transform) spectrum analysis on the basis, and timely alarming the subsynchronous oscillation by setting a self-adaptive oscillation duration threshold. The method can quickly and accurately identify the subsynchronous oscillation behavior, can effectively distinguish the noise of oscillation and suspected oscillation, can timely respond to dangerous oscillation behaviors with high sensitive frequency, high energy accumulation and high oscillation divergence, and can ensure that the alarming information is not sent by mistake in a relatively safe state, so the subsynchronous oscillation identification and alarming method based on PMU (phasor measurement unit) measurement has very remarkable advantages.

Description

Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit)
Technical Field
The invention relates to the technical field of power systems, in particular to a sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement unit).
Background
The rapid development of new energy is an important measure for guaranteeing energy safety and coping with climate change in China, wind power and photovoltaic are main ways for utilizing new energy in China, and ten million kilowatt-level wind power/photovoltaic bases are built in the 'three-north' area, and the utilization of extra-high voltage direct current output is a leading form for developing and utilizing renewable energy in China. In a large-scale new energy collection area, a large amount of power electronic equipment needs to be installed in new energy grid connection, and subsynchronous/supersynchronous harmonic components are introduced into a power signal, so that a subsynchronous oscillation phenomenon of a power system is caused. The stability problem of the Hami-Zhengzhou direct current transmitting end wind power base is outstanding, the subsynchronous oscillation phenomenon is more frequent, and 3 660MW thermal power generating units are caused to trip in serious conditions. Therefore, the real-time monitoring of the subsynchronous oscillation is important for the safe and stable operation of the power grid.
Synchronous Phasor Measurement Units (PMUs) have synchronism, rapidity and accuracy, and can realize dynamic real-time monitoring of a power system.
At present, the power system subsynchronous oscillation identification based on PMU needs to perform FFT spectrum analysis on a large number of data points and distinguish according to the frequency and the amplitude of the obtained inter-harmonic component and the continuous occurrence frequency. The method has large calculation amount and long time, and has a space for improving the rapidity of subsynchronous oscillation identification.
Disclosure of Invention
The invention aims to provide a method for identifying and alarming subsynchronous oscillation based on PMU (phasor measurement unit), which can quickly and accurately identify subsynchronous oscillation behaviors and carry out subsynchronous oscillation alarm in time.
The purpose of the invention is realized by the following technical scheme:
a method for identifying and alarming subsynchronous oscillation based on PMU (phasor measurement Unit) measurement phasor comprises the following steps:
adopting a support vector machine algorithm to distinguish the characteristic vectors extracted from PMU (phasor measurement Unit) measured phasor data with subsynchronous oscillation and non-subsynchronous oscillation in a three-dimensional space, thereby training a corresponding classifier;
for data with unknown types newly measured by PMU, inputting the extracted features into a classifier to realize the identification of subsynchronous oscillation and obtain the data in a data window with subsynchronous oscillation behavior;
carrying out FFT spectrum analysis on data in a data window with subsynchronous oscillation behavior, and extracting inter-harmonic component frequency and amplitude; and filtering out inter-harmonic components caused by noise by using the amplitude of the inter-harmonic components and a predetermined amplitude threshold, and judging whether to generate an alarm signal according to the relationship between the frequency of the residual inter-harmonic components and a self-adaptive oscillation duration threshold.
According to the technical scheme provided by the invention, the subsynchronous oscillation is quickly identified based on the algorithm of the support vector machine, on the basis, inter-harmonic components caused by noise are filtered based on FFT spectrum analysis, and the subsynchronous oscillation is timely alarmed by setting a self-adaptive oscillation duration threshold. The method can quickly and accurately identify the subsynchronous oscillation behavior, can effectively distinguish the noise of oscillation and suspected oscillation, can timely respond to dangerous oscillation behaviors with high sensitive frequency, high energy accumulation and high oscillation divergence, and can ensure that the alarming information is not sent by mistake in a relatively safe state, so the subsynchronous oscillation identification and alarming method based on PMU (phasor measurement unit) measurement has very remarkable advantages.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is a schematic diagram of an algorithm of a support vector machine according to an embodiment of the present invention;
fig. 2 is a flowchart of a sub-synchronous oscillation identification and alarm method based on PMU measurement phasor according to an embodiment of the present invention;
FIG. 3 is a graph comparing three characteristics of an oscillating and non-oscillating data window according to an embodiment of the present invention
FIG. 4 is a schematic diagram of two types of support vectors and interfaces in a training sample according to an embodiment of the present invention
FIG. 5 is a schematic diagram illustrating the variation of the amplitude percentage of inter-harmonic components with time and the corresponding amplitude threshold according to an embodiment of the present invention;
FIG. 6 is a schematic diagram illustrating the percentage change in the amplitude of the inter-harmonic component after being subjected to an amplitude threshold according to an embodiment of the present invention;
FIG. 7 is a graph of energy accumulation and weighted sum of energy accumulation over time according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
With the gradual deepening of the application of the big data technology in the power system, the common characteristic of subsynchronous oscillation is extracted from the massive PMU measurement data, and the artificial intelligence method is utilized for learning, so that the subsynchronous oscillation can be rapidly identified and alarmed. Therefore, the invention provides a sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit) phasor measurement and a support vector machine algorithm in an artificial intelligence method.
Before describing the method provided by the present invention, first, a relevant description is made for the support vector machine algorithm.
The support vector machine algorithm is a machine learning algorithm which can be used for classification, and is a supervised learning algorithm. Given a set of labeled training samples: d { (x)1,y1),(x2,y2),…,(xm,ym)},yi∈{-1,+1}，xiIs an n-dimensional vector and m is the number of samples. The basic idea is to find a partition hyperplane in a sample space based on a training set D, and separate sample points of different classes. As shown in FIG. 1, there are multiple partition hyperplanes satisfying the condition, and the idea of the support vector machine algorithm is to search two kinds of points in n-dimensional spaceThe support vector points in the method enable the sum of the distances from the support vector points of different categories to the partition hyperplane to be maximum, and an interface which has the strongest robustness and is positioned in the middle of the two types of training samples is obtained, so that the method has the strongest adaptability to fresh samples. In sample space, the partition hyperplane is described by the following linear equation:
ωTx+b＝0
where ω is (ω)1；ω2；…；ωd) Determining the direction of the hyperplane for the normal vector; b is a displacement term, and determines the distance between the hyperplane and the origin. The partition hyperplane may be determined by the normal vector ω and the displacement b.
In order to find the interface with the strongest robustness, a conditional extreme value equation is established, namely the basic model of the support vector machine is as follows:
Figure BDA0002404988480000031
s.t.yi(ωTxi+b)≥1,i＝1,2,…,m
its "dual problem" can be obtained using the lagrange multiplier method. Each constraint described above adds a Lagrangian multiplier αiThe lagrange function for this problem is found to be > 0:
Figure BDA0002404988480000032
wherein α ═ (α)1；α2；…；αm)。
Solving a dual problem:
Figure BDA0002404988480000033
Figure BDA0002404988480000034
αi≥0,i＝1,2,…,m
after solving alpha, obtaining omega and b to obtain a model:
Figure BDA0002404988480000041
the above process needs to satisfy the KKT condition, requirement:
αi≥0；
yif(xi)-1≥0；
αi(yif(xi)-1)＝0.
for any training sample (x)i,yi) Always has alpha i0 or yif(xi) 1. If α isi0, then the sample will not appear in the summation of the formula f (x), i.e. there will be no effect on f (x); if α isiIf > 0, then y must be presentif(xi) The corresponding sample point is located on the maximum interval boundary, and is a support vector. This shows an important property of the support vector machine: after training is completed, most training samples do not need to be reserved, and the final model is only related to the support vectors.
If the training sample is not linearly separable, that is, a partition hyperplane cannot be found in the n-dimensional original sample space to correctly partition the two types of points in the training sample, the sample needs to be mapped from the original space to a feature space with a higher dimension, so that the sample is linearly separable in the feature space. The process of mapping from the low dimension to the high dimension is often accomplished by choosing a kernel function. The kernel function implicitly defines a high-dimensional feature space, and the selection of the kernel function determines whether the samples can be linearly separable in the feature space, thereby affecting the classification performance of the support vector machine.
In real-world tasks, it is often difficult to determine a suitable kernel function such that the training samples are linearly separable in feature space, and over-emphasizing linear separability may result in overfitting. To alleviate this problem, the concept of "soft interval" was introduced, allowing the support vector machine to make errors on a small number of samples.
As shown in fig. 2, a sub-synchronous oscillation identification and alarm method based on PMU measurement phasor according to an embodiment of the present invention mainly includes the following steps:
step 1, a support vector machine algorithm is adopted to distinguish feature vectors extracted from PMU (phasor measurement Unit) measured phasor data with subsynchronous oscillation and non-subsynchronous oscillation in a three-dimensional space, so that a corresponding classifier is trained.
The preferred embodiment of this step is as follows:
1) taking historical and/or simulated amplitude data of PMU current phasors as training data; setting a certain window length, and performing feature extraction and dimension reduction on data in each window, so that an amplitude sequence with a certain length is represented as a coordinate point in a three-dimensional space; and determining the label of each coordinate point by using a traditional FFT method, and defining: the oscillation type label value is +1, the non-oscillation type label value is-1, the oscillation is subsynchronous oscillation and does not comprise low-frequency oscillation; and for the data in each data window, if the number of continuous lifting points exceeds a certain value, adding a low-frequency mark to the corresponding data window to indicate that a low-frequency component exists in the data window, and if the type of the corresponding data window is oscillation, distinguishing the low-frequency oscillation from the subsynchronous oscillation.
Illustratively, based on data in a certain data window, a continuous ascending and descending point number sequence is formed, if a numerical value larger than 10 exists in the sequence, namely, more than 10 continuous points continuously ascend or descend exist in the original data window, a low-frequency mark is added, otherwise, the low-frequency mark is not added.
2) Selecting a kernel function, searching respective support vectors of coordinate points of two types in a three-dimensional space, determining an oscillating interface and a non-oscillating interface according to a criterion that the sum of distances from the support vectors of different types to the interfaces is maximum, and training based on training data to form the classifier SVM 1.
The method is characterized in that historical actual PMU data and simulation data are taken to train a SVM1, an oscillation recognition test is carried out on actual PMU data on site, and current phasors of a plurality of lines under the oscillation and non-oscillation scenes are taken under different time, place and voltage levels, so that the test accuracy can reach over 98 percent, and the oscillation recognition method is proved to have better universality.
According to the principle of a support vector machine algorithm, based on the common characteristics of current phasor amplitude data in the oscillation and non-oscillation states, the functional relation between the characteristic vectors extracted from historical and simulated PMU data and the categories of the characteristic vectors is learned, the method can be used for carrying out oscillation judgment on new data, and has reasonability and feasibility. By taking the method as a theoretical basis, the subsynchronous oscillation behavior can be rapidly and accurately identified, dangerous oscillation can be further deeply analyzed, and appropriate response measures can be taken.
And 2, inputting the extracted features into a classifier SVM1 for the data with unknown types newly measured by the PMU to realize the identification of subsynchronous oscillation and obtain the data in a data window with subsynchronous oscillation behaviors.
The preferred embodiment of this step is as follows:
1) setting the same window length as that in the classifier training process for data with unknown types newly measured by PMU, and performing feature extraction and dimensionality reduction on the data in each window, so as to represent an amplitude sequence with a certain length as a coordinate point in a three-dimensional space, wherein the step is the same as the implementation mode of the step 1.
2) Adding a low-frequency mark to the data window with the low-frequency component, and directly judging the data window with the low-frequency mark as a non-oscillation type; for the data window without low frequency label, the classifier SVM1 is used to determine the class.
In the embodiment of the invention, according to the steps 1-2, a certain number of continuous PMU current phasor amplitude data points are taken according to the given window length, three types of features are extracted and dimension reduction is carried out on the points, the points are represented as a coordinate point in a 3-dimensional space, the category of the coordinate point is determined, and the coordinate point is used as the representation of whether the oscillation behavior exists in the time period corresponding to the continuous PMU data points.
In the steps 1-2, values in each dimension of the three-dimensional space are respectively as follows: the number of regular points (feature one), the envelope fluctuation index (feature two) and the number of stationary subsequence points (feature three); wherein:
1) the number of regular points is calculated in a manner that includes:
setting amplitude sequences of PMU current phasors in three temporally continuous data windows as A1 and A, A2 respectively, arranging the amplitude sequences in sequence to form an amplitude sequence A ', marking 1 at the current data point if the current data point is increased or unchanged compared with the previous point from the second data point of the amplitude sequence A', and marking-1 at the current data point if the current data point is decreased, thereby constructing a lifting characterization sequence B1;
according to the lifting representation sequence B1, recording m every time m continuous 1 or-1 appear to obtain a continuous lifting point number sequence C1;
finding out a subsequence with the numerical value of the continuous lifting point number sequence C1 showing periodic change, wherein at least 2 data points are arranged in each period; the subsequence segment in the amplitude sequence A' corresponding to the subsequence is a possible oscillation sequence segment, and all points in the subsequence segment are marked;
and counting the number of marked points in the amplitude sequence A, namely the number of regular points of a data window in which the amplitude sequence A is positioned.
2) The calculation method of the envelope fluctuation index comprises the following steps:
taking an amplitude sequence A of the complete PMU current phasor, a last data point of the amplitude sequence A1 and a first data point of the amplitude sequence A2 which are adjacent to the amplitude sequence A in time, taking the last data point of the amplitude sequence A1 as a front adjacent point of a1 st data point of the amplitude sequence A, taking the first data point of the amplitude sequence A2 as a rear adjacent point of the last 1 data point of the amplitude sequence A, traversing each data point in the amplitude sequence A, and adding the current data point into a sequence B2 if the value of the current data point is greater than or equal to the value of the front adjacent point and the rear adjacent point of the current data point; if the value of the current data point is less than or equal to the front adjacent point and the rear adjacent point, adding the current data point into the sequence C2; the final sequences B2 and C2 are length (B2) and length (C2);
calculating the variances D (A), D (B2) and D (C2) of the amplitude sequence A, the sequence B2 and the sequence C2, and calculating the envelope fluctuation index of the data window in which the amplitude sequence A is positioned by the following formula:
Figure BDA0002404988480000061
3) the calculation mode of the number of the stationary subsequence comprises the following steps:
taking an amplitude sequence A of the complete PMU current phasor and a last data point of the amplitude sequence A1 adjacent to the amplitude sequence A in time, taking the last data point of the amplitude sequence A1 as a previous adjacent point of a1 st data point of the amplitude sequence A, traversing each data point in the sequence A, and marking the current data point if the value of the current data point is unchanged compared with the previous data point;
and counting the number of marked points in the amplitude sequence A, namely the number of stable subsequence points of the data window in which the amplitude sequence A is positioned.
Step 3, performing FFT spectrum analysis on data in a data window with subsynchronous oscillation behavior, and extracting an inter-harmonic component amplitude; and filtering out inter-harmonic components caused by noise by using the amplitude of the inter-harmonic components and a predetermined amplitude threshold, and judging whether to generate an alarm signal according to the relationship between the frequency of the residual inter-harmonic components and a self-adaptive oscillation duration threshold.
The method comprises the following two steps: one part is used for filtering noise; and the other part is alarm judgment.
1. Filtering out noise, the preferred embodiment is as follows:
extracting inter-harmonic component amplitudes through FFT (fast Fourier transform) spectrum analysis, and calculating the percentage of each inter-harmonic component amplitude in the fundamental wave (namely amplitude percentage); and applying an amplitude threshold value to a data window with subsynchronous oscillation behavior, and filtering out the inter-harmonic component with the amplitude percentage smaller than the amplitude threshold value as noise. In this way, the non-harmful inter-harmonic components caused by noise with small amplitude percentage can be eliminated, and attention is paid to the inter-harmonic components with large amplitude and high risk.
In the embodiment of the invention, a Support Vector Machine (SVM) 2 is trained based on FFT spectrum analysis. Based on the boundary determined by SVM2 (in this example, the training data is one-dimensional, so the boundary appears as a one-dimensional boundary point), an adaptive amplitude threshold is set, and the preferred embodiment is as follows:
when a new oscillation scene appears, carrying out FFT spectrum analysis on data in a plurality of continuous data windows with oscillation behaviors to obtain amplitude percentages of various inter-harmonic components, using the amplitude percentages as training data, and training a support vector machine learner; the data categories include: a category of interest, the tag value is + 1; without concern for the class, the tag value is-1; the criteria for the categories are:
the inter-harmonic components with amplitude percentages below N% are directly labeled as don't care classes.
The inter-harmonic component with the amplitude percentage higher than M% is directly marked as a class needing attention; wherein N < M.
For inter-harmonic components which do not meet the first two criteria, if the amplitude percentage of the inter-harmonic components is higher than K times of the average value of all the amplitude percentages of the inter-harmonic components in the same time window, marking the inter-harmonic components as needing attention; otherwise, marking as not needing attention; wherein K is a natural number.
Based on each amplitude percentage data of a given label, a support vector machine learning device is trained to learn the distribution rule of the amplitude percentage data of the class needing attention and the class needing no attention, so that a demarcation point of the two classes of data on a one-dimensional axis is obtained, and the value of the demarcation point is an amplitude threshold value for a certain line under a certain scene.
That is, by collecting a large amount of magnitude percentage data and tagging them with a "focus required class" or a "focus not required class". The support vector machine learner is then trained using these labeled data. The learner SVM2 is trained to distinguish the labeled data in one dimension (axis) by determining a boundary (point) through learning the data. The value of this point is the demarcation point of the data of the two types, namely the type needing attention and the type needing no attention, namely the amplitude threshold value, namely, the value is higher than the value needing attention and the value is lower than the value needing no attention.
As an example, it may be provided that: n is 0.2, M is 1, and K is 5.
In the above scheme, when a new oscillation mode occurs, in order to make the training of the amplitude threshold more effective, when both the attention and non-attention data points exceed a certain value, the support vector machine learner SVM2 is updated in time. Through field mass data tests, the amplitude threshold value fluctuates between 0.4% and 1%.
2. And (4) judging the alarm, wherein the preferred embodiment is as follows:
setting a self-adaptive oscillation duration threshold according to three criteria of sensitive frequency, oscillation divergence speed and energy accumulation, and judging whether an alarm signal is generated or not according to the relation between the inter-harmonic component frequency and the oscillation alarm duration threshold, wherein the steps are as follows:
determining an energy accumulation reference threshold of a single inter-harmonic component according to the inter-harmonic component frequency; if a certain inter-harmonic component is located in a set sensitive frequency range, the energy accumulation reference threshold of the corresponding inter-harmonic component is lower than the threshold of other inter-harmonic components; in addition, an alarm threshold is set for the cumulative weighted sum of the energies of the inter-harmonic components at the same time, wherein inter-harmonic components in the sensitive frequency range are given greater weight than inter-harmonic components in the non-sensitive frequency range.
In the embodiment of the invention, the frequency complementary with the modal frequency of the generator set can be obtained according to the modal frequency of the generator set, and the frequency range near the complementary frequency is set as the sensitive frequency band of the inter-harmonic.
Illustratively, the modal frequency of the generator set is 22Hz, the characteristic frequency complementary with the modal frequency of the power system side is 28Hz, and the sensitive frequency range can be set to be 26.5Hz to 29.5Hz by considering errors. For inter-harmonic components in the non-sensitive frequency range, the reference threshold for energy accumulation is set to 80; in the inter-harmonic component in the sensitive frequency range (26.5Hz to 29.5Hz), the reference threshold for energy accumulation is correspondingly lowered and set to 40. An alarm threshold 200 is set for the energy accumulation weighted sum of the interharmonic components existing at the same time, wherein the energy accumulation of the interharmonic component in the sensitive frequency range is assigned with a weight value of 2, and the energy accumulation of the interharmonic component in the non-sensitive frequency range is assigned with a weight value of 1.
For the continuously occurring inter-harmonics in a certain frequency range, integrating the amplitude percentage of the inter-harmonics with time to serve as an energy accumulated value in a period of time; measuring the average change rate of the energy accumulation value and the energy accumulation weighted sum value of each inter-harmonic component in a previous period of time at intervals of a short period of time, and setting different thresholds (usually setting a plurality of larger thresholds) on the average change rate from small to large according to the gradient in advance; if the change rate is detected to be increased at a certain moment and exceeds the threshold value of a certain gradient, the oscillation divergence speed is high, from the next energy accumulation measuring moment, the energy accumulation threshold value is also reduced according to the corresponding gradient on the basis of the set reference threshold value, so that the alarm speed is accelerated, namely the duration threshold value is reduced.
If at a certain time, the energy accumulation or weighted sum of energy accumulation of a certain inter-harmonic component exceeds the threshold value of the corresponding time, the time is recorded. And (4) solving the time interval from the oscillation starting moment to the moment to obtain a duration threshold value, and sending an alarm signal at the same time.
Experiments also verify the effects of the above scheme of the embodiment of the present invention.
1. And (5) feature extraction testing.
And measuring current phasor data sections by using continuous PMUs of a plurality of different lines near a Xinjiang power grid Hami transformer, a smoke pier transformer and a Tianshan converter station. The window length is taken to be 1s, the sliding step length of the data window is taken to be 1s, and data in the data windows have no intersection. The data segments totally fill 1450 complete data windows, and the data in 825 data windows are tested to be in an oscillation state and the data in 625 data windows are tested to be in a non-oscillation state by a traditional FFT method, so that corresponding class labels are given. The data in each data window is subjected to three feature extractions, and the result is shown in fig. 3. In fig. 3, the three parts (a), (b) and (c) correspond to the first feature, the second feature and the third feature mentioned in step 2; the left and right parts of the dotted line division correspond to the oscillating data window and the non-oscillating data window, respectively.
As can be seen from fig. 3, the three features have better distinguishing capability between the oscillation state and the non-oscillation state, and the numerical difference is more obvious in different states as a whole, for example, the value of the feature one is larger in the oscillation state, and the values of the features two and three are larger in the non-oscillation state. Although the three features have slight data confusion, namely classification errors with a smaller probability, the classifier trained by the joint action of the three features can be improved in accuracy compared with the classifier trained by a single feature by combining three different angles.
2. And (5) cross validation testing.
And (3) taking 25 outgoing line data with different voltage levels near the Hami change and smoke mound change of the Sinkiang power grid at the same time period, and performing characteristic extraction on the PMU current phasor amplitude data. To learn the generalization performance of the classifier trained by using these data, a cross-validation method can be adopted: taking 24 lines of data as a training set, taking the data of the rest 1 line as a test set, and checking the classification precision of the classifier; and taking the 25 lines of line data as a test set in sequence, carrying out oscillation identification test, and averaging test results. The data window lengths with different lengths and different feature combinations are selected for cross validation tests, the results are shown in table 1, and the highest accuracy can reach more than 98%.
Figure BDA0002404988480000091
TABLE 1 Cross validation test accuracy for different window lengths and feature combinations
3. Classifier training
The support vector machine classifier was trained with 25 lines of data at the same time used for the cross-validation test. In this case, the distribution range of the extracted feature vector in the 3-dimensional space may be limited due to a small amount of data. In view of the method for determining the interface of the support vector machine, if the training samples near the interface are insufficient, the interface is easy to generate deviation. Therefore, in order to make the classification boundary more accurate, part of simulation data is added according to the shortage of the distribution range of the training data.
If the 1s window length is selected, for a feature vector (corresponding to a PMU current phasor amplitude data as a stable straight line, namely an ideal amplitude constant current signal) with a feature-one value close to 0 and a feature-three value close to 100, the type of a data window corresponding to the feature vector can be definitely judged to be non-oscillation according to the values and physical meanings of the three features. Such a data window does not appear in the training data, but when some kernel functions such as gaussian kernels are used, the feature vectors are mapped to be close to the separation interface, so that the influence on determining the classification hyperplane is large, and therefore, the artificial supplement is needed.
And (3) training a support vector machine classifier together by using the actual field historical data and a small part of simulation data, and determining a good interface to be used for judging the category of the fresh data. If the combination of the first feature and the third feature is adopted and the window length is set to be 1s, a computer is used for searching support vectors of different categories and determining a boundary surface as shown in the figure 4.
4. Generalization test
The support vector machine classifier was trained using 25 lines of historical data at the same time as used for the cross-validation test, and partial simulation data. In order to verify the recognition effect of the classifier on unseen oscillation and non-oscillation data of other modes, different scenes in different time periods and current phasors on outgoing lines of other transformer substations such as a balcony and a Chinese ephedra trench are additionally taken for carrying out oscillation recognition testing. The generalized test considers factors such as different time, different lines, different voltage levels, oscillation frequency, different oscillation scenes, single oscillation frequency, multiple oscillation frequencies and the like, the test result is shown in table 2, the accuracy rate can reach more than 98%, and the oscillation identification method is proved to have better universality.
Figure BDA0002404988480000101
TABLE 2 generalization test accuracy under different window lengths and combinations of features
From cross-validation testing of training data and generalization testing of large amounts of new data, the following conclusions can be drawn:
(1) under different window lengths and feature combinations, the SVM (support vector machine) classifier obtained by training has better classification accuracy. The classifier accuracy of the single-feature training is slightly low, and under a certain condition, the effect of improving the accuracy can be obtained by combining multiple features.
(2) The classifier accuracy trained from the first feature increases as the window length increases as a whole; the classifier accuracy trained by the second and third features decreases as the window length increases as a whole. The feature-by-feature classification effect is optimal among the three features, and if the window length is increased, a higher accuracy rate can be achieved by only adopting the SVM classifier trained by the feature-by-single feature.
(3) Considering that increasing the window length delays the oscillation recognition time, the amount of calculation is increased; and oscillation can be judged from multiple angles by adopting multiple characteristics, each characteristic has an interference form with poor adaptability, and other characteristics can be complemented with the interference form, so that the performance of the SVM classifier for identifying oscillation can be optimized by selecting the characteristics with short window length and multiple characteristics on the basis of a larger data range.
(4) The generalization test result shows that the classifier trained by utilizing data in a certain period still has higher classification accuracy on unseen new data and good generalization performance. The classifier has better adaptability to the oscillation of different time, different lines, different voltage levels, oscillation frequency, different oscillation scenes, single oscillation frequency or multiple oscillation frequencies.
(5) Further study on the point of classification error shows that the oscillation error is judged to be non-oscillation and the non-oscillation error is judged to be oscillation. The main reason for these two situations is noise interference, so that the data characteristics with oscillation are not obviously misjudged as non-oscillation, or inter-harmonic components are caused to cause the non-oscillation data to be mistaken as oscillation by the classifier; in addition, a portion of the data window may contain both oscillatory data and non-oscillatory data, and the classifier may tend to select the class to which more of the data belongs. For the solution of these problems: on one hand, most points of classification errors exist in an isolated mode, according to the oscillation duration under the general condition and the probability statistics principle, if a certain data window judged to be in oscillation (non-oscillation) exists among a plurality of continuous non-oscillation (oscillation) data windows in an isolated mode, as the accuracy of the classifier is more than 98%, the classifier has great confidence that the window type judgment error exists, and can be corrected; on the other hand, for the problem that noise causes inter-harmonic components, such inter-harmonic components can be filtered out through a subsequent amplitude threshold action process.
5. Classification revision test
According to the category correction thought described in the generalization test conclusion (5), that is, most points of classification errors exist in isolation, according to the oscillation duration under general conditions and the principle of probability statistics, if a certain data window which is judged to be oscillating (non-oscillating) exists in isolation among a plurality of continuous non-oscillating (oscillating) data windows, as the accuracy of the classifier is up to more than 98%, the window category judgment error is considered with great confidence, and the window category can be corrected.
Taking the window length as 1s, redoing the generalization test experiment under various feature combinations, and adding a category correction link: if more than three non-oscillation data windows are arranged before and after a certain oscillation data window respectively, or more than three oscillation data windows are arranged before and after the non-oscillation data window respectively, the correction condition is met, and the data window type is changed into the reverse type of the current type. The accuracy before and after correction is compared as follows:
Figure BDA0002404988480000121
TABLE 3 comparison of generalization test accuracy before and after class correction
As can be seen from Table 3, the accuracy can be improved significantly by adding the category correction link, and the problem of misjudgment of a considerable part of data window categories is solved
6. Low frequency oscillation discrimination test
The invention aims to identify and alarm subsynchronous oscillation, and the low-frequency oscillation has other characteristics which are very similar to subsynchronous oscillation except frequency. Therefore, based on the difference of the continuous ascending and descending point number sequence, a low-frequency mark is introduced. The data window with the low-frequency mark is directly judged to be non (subsynchronous/supersynchronous) oscillation, and the data window class without the low-frequency mark is judged by a support vector machine classifier. Adding low-frequency oscillation components with different sizes into field oscillation and non-oscillation data of the Xinjiang power grid to perform oscillation identification, wherein the ideal result is as follows under two possible conditions:
subsynchronous oscillation and low-frequency oscillation exist simultaneously: the support vector machine classifier judges that the oscillation is (subsynchronous/supersynchronous) and has no low-frequency mark;
there is only low frequency oscillation: the support vector machine classifier judges that the vibration exists, but has a low-frequency mark; or the support vector machine judges that the support vector machine is not oscillatory and has or does not have a low-frequency mark.
By adding a plurality of sine wave components with different amplitudes under the frequencies of 0.5Hz, 1Hz, 3Hz and the like to a large amount of data, the conclusion is as follows:
when subsynchronous oscillation and low-frequency oscillation exist at the same time, the low-frequency oscillation can be correctly judged when any frequency within 3Hz and the amplitude value is lower than that of the subsynchronous oscillation, namely, the data window can be divided into oscillation types, and the data window cannot be judged to be non-oscillation due to the existence of the low-frequency mark. But when the amplitude of the low frequency oscillations exceeds the subsynchronous oscillations, the low frequency signature has a small probability to work, thereby making the classification wrong.
(2) When only low-frequency oscillation exists, the low-frequency oscillation is at any frequency within 3Hz, and the amplitude of the low-frequency oscillation is higher than the average amplitude of the noise existing at the same time, a correct judgment conclusion can be obtained, namely, the support vector machine classifier classifies the data window into a non-oscillation category or the window is marked with low frequency. However, when the amplitude of the low-frequency oscillation is small, even lower than the average amplitude of the simultaneous noise, the data window has a small probability of having no low-frequency mark, and the support vector machine judges the oscillation, so that the classification is wrong.
7. Amplitude threshold test
The method comprises the steps of determining amplitude threshold values of a plurality of oscillation scenes of a Xinjiang power grid, respectively training classifiers for different lines when a new oscillation scene appears, wherein the obtained amplitude threshold values are different due to different amplitude percentages of oscillation components of different lines. Based on simulation experiments on a large amount of data of the Xinjiang power grid, the amplitude threshold value fluctuates between 0.4% and 1%. Fig. 5 shows the variation of the amplitude percentage of the harmonic component between the frequencies with time and the corresponding amplitude threshold value during the primary oscillation process of a certain outlet of the Hami transformer. To reflect this process, 100 consecutive data windows were taken, 12 inter-harmonic components in each window, and the variation of these 12 inter-harmonic components over the 100 consecutive data windows was sequentially spread out in fig. 5 for a total of 1200 data windows. It can be seen from fig. 5 that setting the amplitude threshold has a filtering effect on inter-harmonic components caused by noise.
8. Duration threshold test
The support vector machine classifier is used for identifying certain oscillation of a certain outgoing line of the Hami transformer, and after inter-harmonic components caused by noise are filtered through an amplitude threshold value, the residual inter-harmonic components with larger amplitude percentages are shown in figure 6. In fig. 6, the sections (a) to (e) are a weighted sum of 34Hz inter-harmonic component, 10Hz inter-harmonic component, 20Hz inter-harmonic component, 30Hz inter-harmonic component, and all harmonic components in this order.
The amplitude percentages of the inter-frequency harmonics are integrated over time to obtain an energy accumulation and their weighted sum is shown in fig. 7. In part (a) of fig. 7, the inter-harmonic components are distinguished from each other by a horizontal line shown on the right side, and are 30Hz inter-harmonic component, 34Hz inter-harmonic component, 20Hz inter-harmonic component, and 10Hz inter-harmonic component in this order from top to bottom; (b) and part is a weighted sum of all harmonic components. It can be seen that after 136s, the energy accumulation of the 30Hz inter-harmonic component is increased quickly, and at the moment, the energy accumulation threshold of the component can be reduced, and the alarm speed is accelerated; further, since the inter-harmonic components of a plurality of frequencies appear after 136s, the rate of increase of the energy accumulation weight sum thereof is also fast, and the threshold value for the energy accumulation weight sum is also lowered. It is also emphasized that if a certain inter-harmonic component, for example, 20Hz inter-harmonic component, is located in the sensitive frequency range, the reference threshold of the energy accumulation is lower than the reference threshold of the energy accumulation of 10Hz, 30Hz, 34Hz inter-harmonic components, and has a larger weight when the energy accumulation is weighted and summed.
Through the above description of the embodiments, it is clear to those skilled in the art that the above embodiments can be implemented by software, and can also be implemented by software plus a necessary general hardware platform. With this understanding, the technical solutions of the embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a usb disk, a removable hard disk, etc.), and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods according to the embodiments of the present invention.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (6)

1. A method for identifying and alarming subsynchronous oscillation based on PMU (phasor measurement unit) measurement phasor is characterized by comprising the following steps:
adopt support vector machine algorithm, distinguish the eigenvector that extracts in PMU measurement phasor data to having subsynchronous oscillation and not having subsynchronous oscillation in three-dimensional space to train out corresponding classifier, include: taking historical and/or simulated amplitude data of PMU current phasors as training data; setting a certain window length, and performing feature extraction and dimension reduction on data in each window, so that an amplitude sequence with a certain length is represented as a coordinate point in a three-dimensional space; and determining the label of each coordinate point by using a traditional FFT method, and defining: the oscillation type label value is +1, the non-oscillation type label value is-1, and the oscillation is subsynchronous oscillation; if the number of continuous lifting points exceeds a certain value, a low-frequency mark is added to the corresponding data window to indicate that low-frequency components exist in the data window, and if the type of the corresponding data window is oscillation, low-frequency oscillation and subsynchronous oscillation are distinguished; selecting a kernel function, searching respective support vectors of coordinate points of two types in a three-dimensional space, determining an oscillating interface and a non-oscillating interface according to a criterion that the sum of distances from the support vectors of different types to the interface is maximum, and training based on training data to form a classifier;
for data with unknown types newly measured by PMU, inputting the extracted features into a classifier to realize the identification of subsynchronous oscillation and obtain the data in a data window with subsynchronous oscillation behavior;
carrying out FFT spectrum analysis on data in a data window with subsynchronous oscillation behavior, and extracting inter-harmonic component frequency and amplitude; and filtering out inter-harmonic components caused by noise by using the amplitude of the inter-harmonic components and a predetermined amplitude threshold, and judging whether to generate an alarm signal according to the relationship between the frequency of the residual inter-harmonic components and a self-adaptive oscillation duration threshold.
2. The method according to claim 1, wherein for the data of unknown type newly measured by PMU, the extracted features are input to a classifier, and implementing the subsynchronous oscillation identification and alarm method based on PMU measured phasor comprises:
setting the window length which is the same as that of classifier training for data with unknown type newly measured by PMU, and carrying out feature extraction and dimension reduction on the data in each window, thereby representing the amplitude sequence with a certain length as a coordinate point in a three-dimensional space;
adding a low-frequency mark to the data window with the low-frequency component, and directly judging the data window with the low-frequency mark as a non-oscillation type; and for the data window without the low-frequency mark, judging the category by using a classifier.
3. The method according to claim 2, wherein the values in each dimension in the three-dimensional space are respectively: the number of regular points, the envelope fluctuation index and the number of stationary subsequence points; wherein:
the number of regular points is calculated in a manner that includes:
setting amplitude sequences of PMU current phasors in three temporally continuous data windows as A1 and A, A2 respectively, arranging the amplitude sequences in sequence to form an amplitude sequence A ', marking 1 at the current data point if the current data point is increased or unchanged compared with the previous point from the second data point of the amplitude sequence A', and marking-1 at the current data point if the current data point is decreased, thereby constructing a lifting characterization sequence B1;
according to the lifting representation sequence B1, recording m every time m continuous 1 or-1 appear to obtain a continuous lifting point number sequence C1;
finding out a subsequence with the numerical value of the continuous lifting point number sequence C1 showing periodic change, wherein at least 2 data points are arranged in each period; the subsequence segment in the amplitude sequence A' corresponding to the subsequence is a possible oscillation sequence segment, and all points in the subsequence segment are marked;
counting the number of marked points in the amplitude sequence A, namely the number of regular points of a data window in which the amplitude sequence A is positioned;
the calculation method of the envelope fluctuation index comprises the following steps:
taking an amplitude sequence A of the complete PMU current phasor, a last point of the amplitude sequence A1 and a first point of the amplitude sequence A2 which are adjacent to the amplitude sequence A in time, taking the last data point of the amplitude sequence A1 as a front adjacent point of a1 st data point of the amplitude sequence A, taking the first data point of the amplitude sequence A2 as a rear adjacent point of the last 1 data point of the amplitude sequence A, traversing each data point in the amplitude sequence A, and adding the current data point into a sequence B2 if the value of the current data point is greater than or equal to the value of the front adjacent point and the rear adjacent point of the current data point; if the value of the current data point is less than or equal to the front adjacent point and the rear adjacent point, adding the current data point into the sequence C2; the final sequences B2 and C2 are length (B2) and length (C2);
calculating the variances D (A), D (B2) and D (C2) of the amplitude sequence A, the sequence B2 and the sequence C2, and calculating the envelope fluctuation index of the data window in which the amplitude sequence A is positioned by the following formula:
Figure FDA0002977867210000021
the calculation mode of the number of the stationary subsequence comprises the following steps:
taking an amplitude sequence A of the complete PMU current phasor and a last point of the amplitude sequence A1 adjacent to the amplitude sequence A in time, taking the last data point of the amplitude sequence A1 as a previous adjacent point of a1 st data point of the amplitude sequence A, traversing each data point in the amplitude sequence A, and marking the current data point if the value of the current data point is unchanged compared with the previous point;
and counting the number of marked points in the amplitude sequence A, namely the number of stable subsequence points of the data window in which the amplitude sequence A is positioned.
4. The method of claim 1, wherein the filtering out noise-induced inter-harmonic components using harmonic component amplitudes and predetermined amplitude thresholds comprises:
calculating the percentage of the amplitude of each inter-harmonic component in the fundamental wave, namely the amplitude percentage;
and applying an amplitude threshold value to a data window with subsynchronous oscillation behavior, and filtering out the inter-harmonic component with the amplitude percentage smaller than the amplitude threshold value as noise.
5. A method for identifying and alarming subsynchronous oscillation based on PMU measured phasor according to claim 1 or 4, characterized in that the amplitude threshold is determined by a support vector machine learner as follows:
carrying out FFT spectrum analysis on data in a plurality of continuous data windows with oscillation behaviors to obtain amplitude percentages of various inter-harmonic components as training data to train a support vector machine learner; the data categories include: a category of interest, the tag value is + 1; without concern for the class, the tag value is-1; the criteria for the categories are:
inter-harmonic components with amplitude percentages below N%, labeled as don't care classes directly;
the inter-harmonic component with the amplitude percentage higher than M% is directly marked as a class needing attention; wherein N < M;
for inter-harmonic components which do not meet the first two criteria, if the amplitude percentage of the inter-harmonic components is higher than K times of the average value of all the amplitude percentages of the inter-harmonic components in the same time window, marking the inter-harmonic components as needing attention; otherwise, marking as not needing attention; wherein K is a natural number;
based on each amplitude percentage data of a given label, a support vector machine learning device is trained to learn the distribution rule of the amplitude percentage data of the class needing attention and the class needing no attention, so that a demarcation point of the two classes of data on a one-dimensional axis is obtained, and the value of the demarcation point is an amplitude threshold value for a certain line under a certain scene.
6. The method for identifying and alarming subsynchronous oscillation based on PMU measured phasor according to claim 1, characterized in that the duration threshold of oscillation alarm is determined according to three criteria of sensitive frequency, oscillation divergence speed and energy accumulation, and whether an alarm signal is generated or not is judged according to the relationship between the inter-harmonic component frequency and the duration threshold of oscillation alarm, comprising the following steps:
determining an energy accumulation reference threshold of a single inter-harmonic component according to the inter-harmonic component frequency; if a certain inter-harmonic component is located in a set sensitive frequency range, the energy accumulation reference threshold of the corresponding inter-harmonic component is lower than the threshold of other inter-harmonic components; in addition, an alarm threshold value is set for the energy accumulation weighted sum of the inter-harmonic components at the same time, wherein the inter-harmonic components in the sensitive frequency range are given more weight than the inter-harmonic components in the non-sensitive frequency range;
for the continuously occurring inter-harmonics in a certain frequency range, integrating the amplitude percentage of the inter-harmonics with time to serve as an energy accumulated value in a period of time; measuring the average change rate of the energy accumulation value and the energy accumulation weighted sum value of each inter-harmonic component in a previous period of time at intervals of a short period of time, and setting different thresholds for the average change rate from small to large according to the gradient in advance; if the change rate is detected to be increased at a certain moment and exceeds a threshold value of a certain gradient, the oscillation divergence speed is high, from the next energy accumulation measuring moment, the energy accumulation threshold value is also reduced according to the corresponding gradient on the basis of the set reference threshold value, so that the alarm speed is accelerated, namely the duration threshold value is reduced;
if the energy accumulation or the energy accumulation weighted sum of certain inter-harmonic component exceeds the threshold value of the corresponding moment at a certain moment, recording the corresponding moment; and (4) solving the time interval from the oscillation starting moment to the corresponding moment to obtain a duration threshold value, and sending an alarm signal at the same time.
CN202010158721.0A 2020-03-09 2020-03-09 Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit) Active CN111398679B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010158721.0A CN111398679B (en) 2020-03-09 2020-03-09 Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010158721.0A CN111398679B (en) 2020-03-09 2020-03-09 Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit)

Publications (2)

Publication Number Publication Date
CN111398679A CN111398679A (en) 2020-07-10
CN111398679B true CN111398679B (en) 2021-05-25

Family

ID=71428649

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010158721.0A Active CN111398679B (en) 2020-03-09 2020-03-09 Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit)

Country Status (1)

Country Link
CN (1) CN111398679B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112557753A (en) * 2020-11-05 2021-03-26 许继集团有限公司 Online detection method for low-frequency oscillation of Hanning self-convolution window in frequency division band
CN112611968A (en) * 2020-11-09 2021-04-06 广东电网有限责任公司广州供电局 Phasor measurement subsynchronous oscillation detection method based on decision tree
CN112505452B (en) * 2020-11-25 2024-05-24 东南大学 Wide-band oscillation monitoring method for wide-area system
CN112464848B (en) * 2020-12-07 2023-04-07 国网四川省电力公司电力科学研究院 Information flow abnormal data monitoring method and device based on density space clustering
CN116565862B (en) * 2023-07-10 2023-09-15 华北电力科学研究院有限责任公司 Control method and device of subsynchronous oscillation suppression device

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104300557A (en) * 2014-10-14 2015-01-21 国家电网公司 Electric power system active power oscillation type discriminating method based on support vector machine
CN105226637A (en) * 2015-09-01 2016-01-06 三峡大学 A kind of low-frequency oscillation dispatch control method of the discrimination method that hives off based on vibration
CN105223418A (en) * 2015-09-22 2016-01-06 清华大学 The measuring method of subsynchronous and supersynchronous harmonic phasor and measurement mechanism
CN106841778A (en) * 2016-12-28 2017-06-13 国电南瑞科技股份有限公司 The processing method of the subsynchronous and supersynchronous harmonic parameters realized based on PMU
CN109711435A (en) * 2018-12-03 2019-05-03 三峡大学 A kind of support vector machines on-Line Voltage stability monitoring method based on genetic algorithm
CN110231514A (en) * 2019-06-26 2019-09-13 北京四方继保自动化股份有限公司 A kind of synchronous phasor measuring method suitable for wideband measurement

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101458286B (en) * 2008-12-30 2011-07-13 国家电网公司 Phasor measuring set based on in situ spectroscopic analysis and applied wide domain measuring system
CN102411118B (en) * 2011-12-01 2013-06-26 武汉华中电力电网技术有限公司 Method for judging position of disturbance source for forced power oscillation in regional interconnected power grid
CN104821579B (en) * 2015-04-24 2018-10-23 中国电力科学研究院 A kind of sub-synchronous oscillation method for monitoring and analyzing based on current conversion station electric signal

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104300557A (en) * 2014-10-14 2015-01-21 国家电网公司 Electric power system active power oscillation type discriminating method based on support vector machine
CN105226637A (en) * 2015-09-01 2016-01-06 三峡大学 A kind of low-frequency oscillation dispatch control method of the discrimination method that hives off based on vibration
CN105223418A (en) * 2015-09-22 2016-01-06 清华大学 The measuring method of subsynchronous and supersynchronous harmonic phasor and measurement mechanism
CN106841778A (en) * 2016-12-28 2017-06-13 国电南瑞科技股份有限公司 The processing method of the subsynchronous and supersynchronous harmonic parameters realized based on PMU
CN109711435A (en) * 2018-12-03 2019-05-03 三峡大学 A kind of support vector machines on-Line Voltage stability monitoring method based on genetic algorithm
CN110231514A (en) * 2019-06-26 2019-09-13 北京四方继保自动化股份有限公司 A kind of synchronous phasor measuring method suitable for wideband measurement

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
"ＰＭＵ算法动态性能及其测试系统";毕天姝 等;《电力系统自动化》;20140110;第38卷(第1期);第62-66页 *
"基于PMU和改进支持向量机算法的";田雨 等;《中国电力》;20150331;第48卷(第3期);第50-54页 *
"基于同步相量数据的间谐波还原算法";刘灏 等;《电力自动化设备》;20190131;第39卷(第1期);第153-159页 *

Also Published As

Publication number Publication date
CN111398679A (en) 2020-07-10

Similar Documents

Publication Publication Date Title
CN111398679B (en) Sub-synchronous oscillation identification and alarm method based on PMU (phasor measurement Unit)
Hang et al. Fault diagnosis of wind turbine based on multi‐sensors information fusion technology
Moradzadeh et al. Turn-to-turn short circuit fault localization in transformer winding via image processing and deep learning method
CN110400231B (en) Failure rate estimation method for electric energy metering equipment based on weighted nonlinear Bayes
CN111428755A (en) Non-invasive load monitoring method
CN106060008A (en) Network invasion abnormity detection method
Krátký et al. A novel method for detection of covered conductor faults in medium voltage overhead line systems
Guo et al. Fault diagnosis for power system transmission line based on PCA and SVMs
CN113376516A (en) Medium-voltage vacuum circuit breaker operation fault self-diagnosis and early-warning method based on deep learning
CN101738998A (en) System and method for monitoring industrial process based on local discriminatory analysis
Rezaeian Koochi et al. New phasor‐based approach for online and fast prediction of generators grouping using decision tree
CN116383727A (en) Method, system, equipment and medium for identifying coarse errors in power plant system measurement
CN111651932A (en) Online dynamic security assessment method for power system based on integrated classification model
Ravi et al. Detection and classification of power quality disturbances using stock well transform and improved grey wolf optimization-based kernel extreme learning machine
Shanmugapriya et al. Rapid Fault Analysis by Deep Learning-Based PMU for Smart Grid System.
KR102284958B1 (en) Partial discharge position estimation appratus and method
CN108519993A (en) The social networks focus incident detection method calculated based on multiple data stream
CN116680567A (en) Missing state variable filling method and device for wind driven generator
Sun et al. Electricity theft detection method based on ensemble learning and prototype learning
Chase et al. Detection of Damage in OperatingWind Turbines by Signature Distances
CN115577249A (en) Transformer acoustic signal identification method, system and medium with multi-view feature fusion
CN109145968A (en) A kind of power quality event classification method, system, device and readable storage medium storing program for executing
Wilson et al. Uncertainty Quantification of Capacitor Switching Transient Location Using Machine Learning
Na et al. Inferential sensing and monitoring for feedwater flowrate in pressurized water reactors
Jana et al. Structured approach to downsize data administration and improve fault recognition accuracy for large power systems

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
GR01 Patent grant
GR01 Patent grant