CN107123987A - Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM - Google Patents
Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM Download PDFInfo
- Publication number
- CN107123987A CN107123987A CN201710328990.5A CN201710328990A CN107123987A CN 107123987 A CN107123987 A CN 107123987A CN 201710328990 A CN201710328990 A CN 201710328990A CN 107123987 A CN107123987 A CN 107123987A
- Authority
- CN
- China
- Prior art keywords
- power quality
- quality disturbance
- voltage
- mrow
- electrical energy
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- 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
-
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a kind of electrical energy power quality disturbance recognition methods based on on-line training weighed SVM, comprise the following steps：For the different characteristics of common Power Quality Disturbance, the mathematical modeling of disturbing signal is set up；The Generalized Morphological Filters of ADAPTIVE LMS ALGORITHM are constructed based on mathematical morphology, denoising is carried out to Power Quality Disturbance；The Energy distribution of Power Quality Disturbance different frequency range is extracted by wavelet transformation；Extract the fractal dimension of Power Quality Disturbance according to fractal theory, the quantization spy that Energy distribution and fractal dimension collectively form disturbing signal leads and levies index；The time occurred according to electrical energy power quality disturbance assigns noisy data different weights, realizes on-line training and detection, electrical energy power quality disturbance is identified classification.
Description
Technical field
Field is recognized the present invention relates to the quality of power supply, and in particular to a kind of quality of power supply based on on-line training weighed SVM
Disturbance identification method.
Background technology
China's power system has been developed as a multivariable, the time-varying system of multiple target.In recent years, due to power train
Profound change just occurs for power supply, power network and the load of system, and increasing non-linear, impact and uncompensated load are in electric power
Come into operation in system, such as electric furnace arrangement for producing steel, electric railway, power electronic equipment, cause power network to occur such as voltage wave
A series of electrical energy power quality disturbance problems such as dynamic and flickering, harmonic wave, system frequency fluctuation.Thus, power quality problem has caused
Extensive concern.It is accurate to detect, recognize electrical energy power quality disturbance in order to reach the generation of control disturbance, improve the quality of power supply
Type is the important topic of power system research in recent years.
SVMs (Support Vector Machines, SVM) is grown up based on Statistical Learning Theory
A kind of new general learning method, efficiently solve the practical problem such as small sample, high dimension, non-linear, and overcome people
The shortcomings of artificial neural networks learning structure is uncertain and there is local optimum, greatly increases the generalization ability of learning method.
But current document carries out Training Support Vector Machines after feature extraction to electrical energy power quality disturbance, and new disturbance is identified and returned
Class, and new noisy data is not added in machine training and study in time, i.e., no on-line training and detection, influence training
Speed and accuracy of detection.Meanwhile, the training of the characteristic of nearer Power Quality Disturbance to machine is more important, so will
Ascending weight is set to disturbing signal by the sequencing of appearance, construction Weighted Support Vector is to electrical energy power quality disturbance
Engineering practice can be more conformed to recognizing by carrying out detection.Herein for the problem, the comprehensive each method chief constructs first
The Generalized Morphological Filters of ADAPTIVE LMS ALGORITHM, to carrying out de-noising pretreatment containing noisy disturbing signal；Secondly using multiple small
Wave conversion and fractal theory extract the Energy distribution and Local Fractal Dimension of signal different frequency range respectively, with reference to Energy distribution and point
The characteristic quantities such as shape dimension characterize different Power Quality Disturbances, set ascending to disturbing signal by the sequencing of appearance
Weight, construction on-line training Weighted Support Vector electrical energy power quality disturbance is detected and recognized, raising accuracy of identification.
The quality of power supply is the quality of electric energy in power system.Preferable electric energy should be the sine wave of ideal symmetrical.Some
Factor can make waveform deviate symmetrical sine, thus just generate power quality problem.From the strict meaning, the quality of power supply is weighed
Refer mainly to indicate voltage, frequency and waveform.Refer to quality supply from the universal significance quality of power supply, including quality of voltage,
Current quality, power supply quality and power quality.It can be defined as：Cause the electricity of electrical equipment failure or cisco unity malfunction
The deviation of pressure, electric current or frequency, its content includes frequency departure, voltage deviation, voltage fluctuation and flicker, three-phase imbalance, wink
When or transient overvoltage, wave distortion (harmonic wave), voltage dip, interruption, temporarily rise and power supply continuity etc..
(1) power quality problem
Power quality problem includes quality of voltage problem and current quality problem.The quality of power supply is characterized point by wave distortion
Class mainly includes harmonic wave, subharmonic, waveform and sunk and noise etc.；Being characterized classification by frequency spectrum and transient state duration includes arteries and veins
Rush transient state and the vibration major class of transient state two.Dynamic power quality problem is mainly reflected in quality of voltage problem.
1. voltage deviation
Voltage deviation is the general name that voltage falling (Voltage Drop) and voltage rise (voltage protuberance).
2. frequency departure
Requirement the whole network to frequency quality is identical, different not because of user, and there are related rule various countries for this deviation standard
It is fixed.
3. voltage three-phase imbalance
The average value of peak excursion and three-phase voltage that voltage three-phase imbalance shows as voltage exceedes defined standard.
4. harmonic wave and m-Acetyl chlorophosphonazo
Sinusoidal voltage or electric current containing fundamental wave integer multiple frequency are referred to as harmonic wave.Sine containing the non-integral multiple frequency of fundamental wave
Voltage or electric current are referred to as m-Acetyl chlorophosphonazo, and the fractional harmoni less than fundamental frequency falls within m-Acetyl chlorophosphonazo.
5. voltage fluctuation and flicker
Voltage pulsation refers to that the regular variation of the voltage in envelope, or amplitude are usually not more than 0.9~1.1 times
The a series of voltage of voltage range changes at random.Flickering then refers to visual impact of the voltage pulsation to illuminating lamp.
In terms of causing the factor of above-mentioned power quality problem to have following three：
1. the nonlinear problem that power system component is present
The nonlinear problem of power system component mainly includes：The harmonic wave that generator is produced；The harmonic wave that transformer is produced；Directly
The harmonic wave that stream transmission of electricity is produced；Transmission line of electricity (particularly extra high voltage network) is to the amplification of harmonic wave.In addition, also power transformation
Influence of the factors such as shunt capacitor compensation device of standing to harmonic wave.
2. nonlinear-load
In industry and household electricity load, nonlinear load accounts for significant proportion, and this is the master of problem for power system harmonics
Originate.A large amount of non-linear, fluctuations such as rectifier, pressure regulation and converter plant, electric arc furnaces, electric railway, impact and not
Balanced load is widely used, and is distorted the voltage of power network, current waveform or is caused voltage fluctuation and flicker and three
Phase imbalance etc., or even cause system frequency fluctuation, voltage interruption, voltage transient variation and voltage transient, so as to cause the quality of power supply
Decline and severe jamming and pollution are caused to the quality of power supply of power network.
3. electric power system fault and dynamic operation
The failure of Operation of Electric Systems will also result in power quality problem with dynamic operation, such as various short troubles, lightning
When electric shock circuit, artificial maloperation, electric network fault in the change of the working condition of generator and excitation system, failure protecting device
The dynamic operation of startup and power system of power electronic equipment etc. will all to cause the voltage of power network, current waveform to occur abnormal
Become or cause the various power quality problems such as voltage fluctuation and flicker.
(2) Power quality management As-Is analysis
With power network intellectualized reconstruction deepen continuously and new technology application, the electric network composition of China is increasingly perfect,
Electric network performance is continued to optimize, and can meet the primary demand of power grid user.But, also exist not in terms of Power quality management
Foot, according to statistics, caused quality of power supply ratio not up to standard has exceeded 60% the problem of due to Power quality management.The quality of power supply
The subject matter that management is present has at following 5 points.
1. timely substation bus bar voltage automatic regulation mechanism has not been set up.Bus regulation and control are to cause voltage not in time
Out-of-limit main cause.
2. small power supply management system is lacked.The online lack of standardization of small power station can cause transformer station's 35KV and 10KV bus
Voltage out-of-limit.
3. power network unreasonable allocation.For example Urbanization Progress is very fast in recent years, and original suburb, which has been developed rapidly, to be turned into
Urban new center, but power network power supply facilities do not catch up with, and causes line powering radius excessive, directly affects power supply reliability and
Stability.
4. power network lower management low SI.There is data to suggest that, the duration of following voltage out-of-limit at county level or at county level is much big
In higher level parent company.
5. electric network pollution, which is administered, lacks.To there is substantial amounts of nonlinear-load, impact load and imbalance in power network negative in itself
Lotus access system etc., forms serious threat to the quality of power supply, and the improvement to these pollutions is not paid attention to fully also at this stage.
The content of the invention
To solve the above problems, the invention provides a kind of electrical energy power quality disturbance identification based on on-line training weighed SVM
Method.
To achieve the above object, the technical scheme taken of the present invention is：
Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM, comprises the following steps：
S1, the different characteristics for common Power Quality Disturbance, set up the mathematical modeling of disturbing signal；
S2, the Generalized Morphological Filters based on mathematical morphology construction ADAPTIVE LMS ALGORITHM, to Power Quality Disturbance
Carry out denoising；
S3, the Energy distribution by wavelet transformation extraction Power Quality Disturbance different frequency range；Carried according to fractal theory
Take the fractal dimension of Power Quality Disturbance, the quantization spy that Energy distribution and fractal dimension collectively form disturbing signal leads and levies finger
Mark；
S4, the time occurred according to electrical energy power quality disturbance assign noisy data different weights, realize on-line training and inspection
Survey, electrical energy power quality disturbance is identified classification.
Wherein, the mathematical modeling of the disturbing signal of the step S1 includes
1. voltage dip (Voltage sag)
Wherein, 0.1 ＜ a ＜ 0.9 are temporary range of decrease degree, T ＜ t2-t1＜ 8T are temporarily drop duration, t1During to take place
Carve, t2For finish time, T is the cycle；
2. voltage swell (Voltage swell)
Wherein, 1.1<a<1.8 be temporary increasing degree degree, T<t2-t1<8T is temporarily liter duration, t1For moment, t takes place2
For finish time；
3. voltage interruption (Interruption)
Wherein, 0<a<0.1 is interruption amplitude, T<t2-t1<8T is duration of interruption, t1For moment, t takes place2For
Finish time；
4. transient state (Oscillatory transients) is vibrated
Wherein, 0.1 ＜ a ＜ 0.8 are vibration amplitude peak, 5<λ<15 be vibration frequency relative coefficient, 10<p<100 be to shake
Swing attenuation coefficient, T ＜ t2-t1＜ 8T are duration of oscillation, t1For moment, t takes place2For finish time；
5. harmonic wave (Harmonics)
Wherein, 0.05 ＜ a3The ＜ a of ＜ 1,0.055The ＜ a of ＜ 1,0.057＜ 1 is 3 times, 5 times and 7 subharmonic amplitudes, for it
Its subharmonic content is typically smaller, not impact analysis result；
6. voltage pulsation (Voltage fluctuation)
Wherein, UmFor work frequency carrier voltage magnitude, aUmCos λ ω t are frequency-modulated wave sinusoidal voltage waveform, and ω is work frequency carrier
Voltage angular frequency, 0.1 ＜ a ＜ 0.5,0.1<λ<0.5.
Wherein, the Generalized Morphological Filters of the ADAPTIVE LMS ALGORITHM are：
It is One-dimension Time Series to define input signal x (n), and g (n) is structural element, and structural element g (n) length is less than
Signal x (n) length.For two structural element g1 (n) and g2 (n),And g1 (n) length is less than g2
(n) length, then constitute Generalized Morphological open-close, close-opening operation；
Wherein, o represents opening operation, represents closed operation.The output expression formula of so Generalized Morphological Filters is as follows：
Y (n)=α y1(x)+βy2(x) (9)
In formula：α and β are weight coefficients, and alpha+beta=1
Assuming that noise-free signal is Q, then least mean-square error is：
The invention has the advantages that：
(1) this specific implementation constructs the Generalized Morphological Filters based on ADAPTIVE LMS ALGORITHM according to mathematical morphology, to electricity
Energy quality disturbance signal carries out denoising, can improve the speed and precision of electrical energy power quality disturbance identification.
(2) specific time-frequency advantage is shown when handling singular signal using the multiscale analysis of small echo, to the quality of power supply
Disturbing signal is handled, and extracts the Energy distribution of disturbing signal.Meanwhile, based on fractal geometrical theory, the quality of power supply is disturbed
It is dynamic to extract local variance dimension.The two is combined to the characteristic vector for collectively forming Power Quality Disturbance, accelerates supporting vector
The disturbed depth speed of machine.
(3) time occurred according to electrical energy power quality disturbance assigns noisy data different weights, meanwhile, to new disturbance not
Only it is identified, and is enriched training set and continue Training Support Vector Machines, using delta algorithm, can be significantly reduced
On-line training weights demand of the supporting vector algorithm to Installed System Memory, reduces the training time, the real-time of enhancing algorithm modeling.
Brief description of the drawings
Fig. 1 be voltage flicker original signal (on) and de-noising signal (under).
Fig. 2 be harmonic current original signal (on) and de-noising signal (under).
Fig. 3 be voltage swell original signal (on) and de-noising signal (under).
Fig. 4 be voltage dip original signal (on) and de-noising signal (under).
Fig. 5 be voltage interruption original signal (on) and de-noising signal (under).
Fig. 6 be transient oscillation original signal (on) and de-noising signal (under).
Fig. 7 is the simulation waveform detected based on sliding window iteration DFT method.
Fig. 8 is whole detection and classifying and identifying system software flow.
Fig. 9 is system the general frame.
Figure 10 is wireless telecommunications flow chart.
Embodiment
In order that objects and advantages of the present invention are more clearly understood, the present invention is carried out with reference to embodiments further
Describe in detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to limit this hair
It is bright.
Using the adaptive technique based on least-mean-square error algorithm, by progressively modified weight coefficient, input signal is followed the trail of,
Once the statistical property of input signal changes, its energy adjust automatically parameter, make the wave filter that there is more preferable de-noising ability,
And more maintain marginal information and useful information.
It is One-dimension Time Series to define input signal x (n), and g (n) is structural element, and structural element g (n) length is less than
Signal x (n) length.For two structural element g1 (n) and g2 (n),And g1 (n) length is less than g2
(n) length, then constitute Generalized Morphological open-close, close-opening operation.
Wherein o represents opening operation, represents closed operation.The output expression formula of so Generalized Morphological Filters is as follows：
Y (n)=α y1(x)+βy2(x) (9)
In formula：α and β are weight coefficients, and alpha+beta=1
Assuming that noise-free signal is Q, then least mean-square error is：
Fig. 1-Fig. 6 is voltage dip, voltage swell, voltage interruption, vibration transient state, harmonic wave and the disturbing signal of voltage pulsation
The simulation result of Generalized Morphological Filters through ADAPTIVE LMS ALGORITHM.From the point of view of these results, based on ADAPTIVE LMS ALGORITHM
The denoising effect of Generalized Morphological Filters is good.
Feature extraction is carried out to Power Quality Disturbance using wavelet transform and fractal is theoretical common, can be with less
Characteristic quantity characterizes different faults type, accelerates the disturbed depth speed of SVMs.
Harmonics and reactive current detection based on frequency domain
For cycle T=2 π/ω non-sinusoidal current signal i (ω t), if meeting Di Liheli three conditions, then i
(ω t) just can resolve into the Fourier space form as shown in formula (11)：
Wherein：
Again by Euler's formula ejnωt=cos (n ω t)+jsin (n ω t) and e-jnωt=cos (n ω t)-jsin (n ω t),
Convolution (11), can obtain as shown in formula (12)：
N span is expanded to whole real number field simultaneously, then sin (n ω t) and cos (n ω t) is respectively the strange of n
Function and even function, obtain a0As shown in formula (13)：
Convolution (12) and formula (13), shown in the exponential form such as formula (14) for drawing Fourier space：
Assuming thatThenUnderstand, the amplitude of nth harmonic isArgument isAnd becauseWithFor conjugate complex number, deploy when by n from-∞ to+∞
When, it can obtain shown in nth harmonic such as formula (15)：
But specific in the engineering practice of electric railway, the current waveform of the distortion in power network is unknown, therefore can not
By it is equal with some function analytic expression in advance after, then carry out Fourier space and calculated.Therefore, when conventional means are pair
Between continuous periodic current waveform N number of equally spaced sampled point inserted in cycle T carry out interval sampling, then sampled value is turned
Become a series of data signals, quick Fourier analysis is carried out using computer.So for Fu of above-mentioned continuous function
In leaf-size class number must be converted to the approximated Fourier series of corresponding discrete series, thus continuous function i (t) is converted into discrete sequence
Arrange { ik}.AndDt=Δs t=τ=T/N.In conjunction withIt can obtain：
Wherein n=1,2 ... ..N-1；
Wherein n=1,2 ... ..N-1；
It can obtain shown in discrete Fourier transform formula such as formula (16)：
Wherein n=1,2 ... ..N-1 (16)
To improve the real-time of detection, sliding window iteration thought is introduced.Purpose is so that anAnd bnCalculate need not from (k=0) that
Any starts to count, but proceeds by calculating from the data point of last samples.The renewal speed of data is so accelerated, is also made
Obtain computation amount.Improved anAnd bnAs shown in formula (17) and formula (18)：
In formula, NcurrentRepresent newest sampled data points；I (k τ) represents sampled point NcurrentThe sampling in k cycle before
Value.
According to above-mentioned, it is assumed that fundamental current component i1(t τ)=a1cos(ωtτ)+b1Sin (ω t τ), a1And b1Summation meter
Calculation can be reduced to as shown in formula (19) and formula (20)：
Formula (19) and formula (20) show that the corresponding fundamental current content of t is sampled point NcurrentK cycle before
Sampled value sum subtracts NcurrentPrevious value, along with NcurrentCurrent value.New sampled value sum after calculating replaces old
Sampled value sum, complete iteration.So improve the renewal speed of data so that the ability enhancing of follow current change.
By taking electric railway as an example：The harmonic content of 3 times, 5 times and 7 times is higher in electric railway.By electric locomotive mould
In distortion current numerical value feeding Workspace in type.(19) and (20) of the mathematical derivation of DFT based on sliding window iteration, are examined
Consider can respectively simultaneously detect 3 times, 5 times and 7 subharmonic.The sample frequency for providing fundamental wave is 6400Hz, it is seen that nth harmonic
Sample frequencySampling number is
Resulting l-G simulation test waveform from Fig. 7 as shown in fig. 7, draw a conclusion：1. DFT detection sides based on sliding window iteration
Method is complete and accurately detected 3 times, 5 times and 7 order harmonic components in distortion current；2. advantage is：What is detected is humorous
Wave error is small, real-time is good；Overall operand is small.3. shortcoming is：The reactive current amount of change can not be detected.Therefore use
The result of this detection method is can not to produce the instruction current of reactive-load compensation so that inverter can only compensate harmonic wave in power network,
But idle it cannot be compensated to what is changed in power network.
The extraction of energy-distributing feature
For arbitrary signal x (t) ∈ L2(R), its multiresolution is：
Wherein：J represents the number of plies decomposed；c0(k) it is scale coefficient；dj(k) it is wavelet coefficient；
It is expressed as follows using the Double-scaling equation of filter form：
Wherein：H (k), g (k) are respectively low pass filter and high-pass filter coefficient, and are met：
G (k)=(- 1)kh(1-k) (24)
The characteristic vector of Energy distribution under using different scale as sample.Energy under each decomposition scale uses following formula
Calculate：
Wherein：I is the yardstick of wavelet decomposition；N is the coefficient sum of detail signal and approximation signal；Exist for approximation signal
Energy distribution on different scale；The Energy distribution for being detail signal on different scale；
If disturbing signal can obtain the characteristic vector of one (l+1) dimension, i.e., by l layers of wavelet decomposition：
Feature=[ED1,ED2,···,EDl,ECl] (27)
Therefore, 6 layers of wavelet decomposition can obtain 7 dimension the feature parameter vectors.
The extraction of fractal dimension
For point setIf can be by n dimension hypercube covering of the individual length of sides of N (r) for r, point set Ω box dimension
Number is：
Because box counting dimension does not reflect the inhomogeneities of geometric object, the box containing one or more points has in above formula
There is same weight, information dimension improves to some extent to this.The probability that note point set Ω falls into length of side r k-th of hypercube is Pk
(r) information content required for, system mode is accurate to r magnitudes is defined as entropy, then：
Then point set Ω information dimension is：
In real-time implementation, variance dimension can be calculated, signal x (t) is discrete, the amplitude increment in time Δ t
Variance is according to σ2Contacted according to following power-law equation and incremental time：
Var[x(t2)-x(t1)]~| t2-t1 2H| (31)
Wherein：H is Hurst indexes.It is distributed in order that point is equally spaced on logarithmic chart, generally selects sequence and variance
Using b as the logarithm at bottom, i.e.,：
Xk=logb(Δtk), Yk=logb Var(Δxk) (32)
One multinomial linear fit method determines the slope s of these points, and this slope calculating formula is as follows.
If analyzed Power Quality Disturbance is with 4096 points of 12800sps sample rates, wherein calculating part side
The window width of difference dimension is taken as at 256 points, every time 128 points of displacement.The electrical energy power quality disturbance being so made up of local variance fractal dimension
The feature image's dimensions of signal are 30.7 dimensions of the energy-distributing feature vector that combined wavelet transformation is extracted, then for same disturbance
Signal can characterize its characteristic with the characteristic vector of 37 dimensions, and this is the data source of SVM training and test.
Fourier transformation, wavelet transformation and fractal dimension extract characteristic quantity jointly
37 D feature vectors of above-mentioned 7 dimension wavelet transformation energy and 30 dimension local variance dimension compositions, characteristic quantity is too many, increases
The complexity calculated is added.Characteristic quantity is extracted using Fourier transformation, wavelet transformation and fractal dimension jointly, meter can be simplified
Calculate.Fourier transformation is carried out to the disturbance waveform after sampling by Matlab programs, fundametal compoment is extracted, phase angle shift, total humorous
The characteristic vector of ripple aberration rate, low-frequency harmonics aberration rate；Two scale wavelet transforms are carried out to disturbing signal using small echo and extract small
Wave system number energy；The box counting dimension composition 6 for extracting the disturbance moment using fractal theory ties up the characteristic quantity of disturbing signal.
In above formula：|Vn[1] | it is the Fourier transform of sample signal；WCnWavelet coefficient for sampled signal per cycle；le
For WCnLength.
Normalized
To improve the generalization ability of model, reducing the time of procedural training, data normalized has been subjected to.This is specific
Implement all data being mapped to [- 1,1] interval, normalization formula is：
In formula, xi' for normalization after value, xmaxAnd xminRespectively initial data xiMaximum and minimum value.Using
Mapminmax function pairs sample data has carried out normalized in Matlab, it is to avoid data area caused greatly training, improves
Disturb the accuracy of classification.
Weighted Support Vector on-line training algorithm
Because power system is a time-varying system, new inputoutput data is continuously available, in order that model can be accurate
The current state of ground reflection system is it is necessary to use new data descriptive model, and the legacy data diminished with current time correlation can
To ignore or shared proportion in modeling should be reduced, therefore, it is interval to set up a modeling data rolled with the time, and keeps
The interval length L is constant.With the operation of system, the data field of data interval is changed over time, the importance of Recent data
It is significantly larger than the importance of early time data.Therefore, using Weighted Support Vector, different errors is assigned to each sample and are punished
Penalty factor, improves disturbed depth precision.
(1) mathematical modeling of Weighted Support Vector
If given training set (xi,yi),xi∈Rn, yi∈ { -1,1 }, i=1,2 ..., 1.One is found based on training set
Decision function f (x)=<w·x>+ b, on the one hand will make the interval between 2 class sample points as big as possible, and it is equivalent to minimize | |
w||2/ 2, on the other hand make wrong point degreeIt is as small as possible, then the problem is converted into the following convex quadratic programming problem of solution.
Wherein：ξ is loss function, and c takes fixed punishment parameter in advance for some, and it embodies the attention journey labelled unjustifiably to sample
Degree.It is identical for different sample punishment in standard SVMs (C-SVM).But in actual applications it is often found that
Some sample importances are big, it is desired to small training error；And the importance of some samples is relatively lower, a certain size is allowed
Training error.That is, each sample should have different error penalty coefficients, can just obtain more accurately prediction effect.According to reality
Border situation is to each training sample xiAssign a weights ri, a weights c is assigned to every class sampley, then weighting supporting vector
Machine (W-SVM) can be described as：
Its Lagrange function is：
Wherein, ai,βiRespectively constrain yi(<w,g(xi)>+b)≥1-ξiLagrange multipliers.
By constituting Lagrange functions, then according to KKT conditions, the dual problem of above formula just can be obtained.
Above-mentioned dual problem is solved, the regression estimates function for obtaining W-SVM is：
In formula：αi、It is Lagrange multiplier；
K(xi, x) it is kernel function；
B is the bias of function model.
(2) the online SMV training algorithms of increment type
Define error function：h(xi)=f (xi)-yi, according to the h (x of each samplei) and parameter θiValue, by training sample
It is divided into 3 subsets：Mistake supporting vector set E, edge support vector set S and reservation sample set R, are defined such as respectively
Under：
E=i | θi=-c ∧ h (xi)≥+ε)∨(θi=+c ∧ h (xi)≤-ε)} (41)
S=i | θi∈[-c,0]∧h(xi)=+ ε) ∨ (θi∈[0,+c]∧h(xi)=- ε) } (42)
R=i | θi=0 ∧ | h (xi)|≤ε} (43)
Wherein, parameter c, ε are respectively the penalty factor and allowable error in SVMs optimization problem.
The training objective of model is by new samples { xc,ycOne of above-mentioned 3 set are added, while making all samples still full
Sufficient KKT conditions.Its method is：When | h (xc)|<During ε, new samples add R collection, original training sample set and θiValue need not change；When
|h(xc) | during >=ε, new samples add S collection or E collection, must ensure θ caused by new samples before addingiWith h (xi) change of value will not
The composition of original 3 set is influenceed, otherwise first original collection must be combined into and be adjusted, i.e., sample is moved.Sample movement can
It is divided into 3 kinds of situations：R collection or E collection are moved to from S collection；S collection is moved to from E collection；S collection is moved to from R collection.Mobile object and side
Formula depends on the mobile new samples θ causedcVariable quantity (Δ θ c values).Each iteration selection makes Δ θcMinimum direction carries out sample
This movement, until new samples add S collection or E collection.
(3) algorithmic characteristic is analyzed
The time complexity of algorithm depend primarily on sample movement number of times and sample movement after parameter renewal.Parameter is more
New time complexity is O (n × ls) × O (Kernel), wherein, n is training sample sum；lsFor supporting vector number；
Kernel represents SVM kernel functions.The time complexity of this specific implementation algorithm in the worst cases is O (n3)×O(Kernel)。
Calculating the computation complexity O (Kernel) of kernel function can be removed by caching nuclear matrix, and cost is increase O (n2)
Space complexity.Therefore, although this specific implementation algorithm O (n in the worst cases3) time complexity it is higher, but consider
The changeability of number of times is moved to sample, under average case, the time complexity of algorithm is O (cn2).In this specific implementation algorithm
In, when new samples are added each original sample it is every kind of be moved to carry out once, therefore, the mobile total degree of sample for O (n ×
3)。
The confidence level of algorithm changes with ε change, if when ε is relatively small, if confidence level is also very high, said
Almost without noise in bright data, on the contrary, the less words of Reliability ratio, illustrate that certain noise may be contained in data.From straight
Said in sight, the relation between ε and confidence level can reflect the number of Noise in data.Certainly, the size of confidence level also and branch
The ε values held in vector machine loss function are relevant, and ε values are smaller, and confidence level should be bigger.In a word, by the adjustment to each parameter,
The content of noise in data can be identified to a certain extent.
Emulation experiment
The electric energy of the type of harmonic voltage, voltage swell, voltage dip, voltage interruption, ringing and voltage flicker etc. 6
Quality signal uses C successively1-C6Represent, electric energy quality signal always produces 600 groups of samples, and every kind of electrical energy power quality disturbance is respectively 100
Group, wherein 40 groups are used to train, 60 groups are used to detect in addition.In order to preferably emulate the disturbance feelings of the real system quality of power supply
Condition, adds the white Gaussian noise that signal to noise ratio is 15db in voltage disturbance.
The mother wavelet decomposed from db4 small echos as Mallat.All signals are all with identical sample rate (256 points/cycle)
Sampling, sampling number is 4096 altogether.6 Scale Decompositions are carried out to signal using Mallat algorithms, and by the energy under different scale
Amount distribution is used as the Part I characteristic vector disturbed.Disturbing signal variance fractal dimension is extracted using adding window displacement method and made
For the Part II characteristic vector of disturbance.The length of disturbing signal is 16 cycles, half period is shifted every time, therefore can be formed
30 variance dimensions, add 7 characteristic vectors of Part I totally 37 dimension.Gaussian radial basis function is selected as kernel function, using based on
The multi-category support vector machines algorithm of two classification carries out on-line training and identification, its recognition result such as table 1 below to characteristic vector
It is shown.In order to compare, experiment condition is constant, selects gaussian radial basis function as kernel function, is carried out using based on standard SVMs
Training and identification, its recognition result are as shown in table 2 below.
Table 1:M-SVM recognition results
Table 1:C-SVM recognition results
As can be seen from the above table, C-SVM is that 96.67%, M-SVM is average to disturbed depth to disturbed depth average accuracy
Accuracy is 97.5%.Therefore online M-SVM can be very good to recognize different types of Power Disturbance, with preferable identification essence
Degree.
Electric energy quality monitoring system
By the running environment and feature of power system, detect Power Quality Disturbance and inevitably include
Noise signal.The presence of noise signal can reduce the accuracy of detection, in the strong occasion of noise signal, or even can cause detection
Failure.Classify again for this reason, it may be necessary to carry out first denoising to Power Quality Disturbance.Overall system design thinking is by data acquisition
After the data normalization that unit is collected, through the Generalized Morphological Filters denoising based on ADAPTIVE LMS ALGORITHM.Denoising
Noisy data send into wavelet character extraction module and variance fractal dimension extraction module simultaneously.The two characteristic vector extracted is sent
The Weighted Support Vector module for entering on-line training realizes that Classification of Power Quality Disturbances is recognized.Whole detection and classifying and identifying system
Software flow is as shown in Figure 8.
Total system uses ARM+DSP Slave Parallel processing system, the disturbance type identification based on SVMs
Concentrated on human-computer interaction function in ARM subsystems, the control to all peripheral hardwares is completed by main frame.Utilize DSP rapid data
Disposal ability complete to extract three-phase voltage signal, the collection of three-phase current signal, denoising, wavelet transformation and point shape feature to
Amount.Data communication between ARM and DSP is realized by a dual port RAM.As shown in Figure 9.
(1) the Hardware Design
1. DSP and ARM core circuit designs
This specific implementation uses ARM+DSP Slave Parallel processing system, the disturbance type based on SVMs
Identification, human-computer interaction function and radio communication function are concentrated in ARM subsystems, and the control to all peripheral hardwares is completed by main frame.
Completed using DSP fast data processing capability to voltage signal, the collection of current signal, denoising and extract characteristic vector.
Data communication between ARM and DSP is realized by a dual port RAM.Dsp chip selects TMS320VC5402 chips, the core
Piece is the fixed-point DSP chip that TI companies need and specially designed for low-power consumption, high-performance；ARM chips selection Samsung companies
ARM9 family chip S3C2420, with reference to corresponding peripheral hardware constitute a complete ARM application system, with small volume, power consumption
It is low, the features such as relative processing capacity is strong, can load and run operating system, realize multi-task scheduling, improve PQD knowledges
Not, the reliability and rapidity of radio communication.
2. data acquisition unit is designed
Data acquisition unit design is:Using small-sized D.C mutual-inductor, 100V, 5A primary voltage, electric current are believed
The weak electric signal between+5V--5V number is converted into, and signal condition is carried out by high-precision operational amplifier, by low pass filtered
After ripple, A/D change-over circuits are sent to.In order to quickly and accurately reflect the quality of power supply of power network, it is desirable to which the partial circuit is necessary
Ensure the very high linearity.The present apparatus selects microminiature, high-accuracy electric current and the voltage changer of Liao Dongsheng companies.This conversion
The device linearity is that phase shift is less than 70 ' after 0.1% compensation, and isolation voltage is up to 2500V, and small volume, lightweight, can directly be welded
On a printed-wiring board.Analog quantity is completed to the conversion of digital quantity from ADS8346 chips.ADS8346 is that TI companies aim at height
A 16 A/D conversion chips of fast synchronous data collection design, are made up of, each 3 switching rates for 250kS/s ADC
ADC has 2 analog input channels, and the analog quantity conversion of 6 passages can be realized simultaneously.
3. radio receiving transmitting module is designed
ZigBee uses IEEE802.15.4 standards, using the shared public frequency 2.4GHz in the whole world, applied to monitoring, control
During network processed, the advantage such as its low cost with highly significant, low power consumption, many, long transmission distance of network node is currently viewed as
Substitute one of wired monitoring and the control most promising technology of network field.2.4GHz less radio-frequency core is supported in the market
The type and quantity of piece are relatively more, mainly there is the chips such as AP1110, nRF24L01, CC1100, CC2420, CC2430.CC2430
Chip using powerful IDE as support, the interactive debugging of internal wiring using defer to IDE IAR industrial standards as
Support, obtain the highly recognition of embedded mechanism.This project selection is highly integrated, low-power consumption, support Zigbee protocol chip
CC2430 completes the design of wireless transceiver circuit.
(2) Design of System Software
1. embedded OS TinyOS
Domestic existing power quality detection system, its data acquisition unit with the data communication between control centre mostly
Carried out by wired mode, bottom communication mostly uses fieldbus (such as RS485, CAN), telecommunication mode
There are optical fiber, power carrier, public network, wire cable etc., bring very big inconvenience to work such as track laying, overhauls of the equipments, build into
This and engineering remain high.The appearance of radio sensing network solves the problem of cable network is present well, and it has very big
Flexibility, it is only necessary to electric power detection zone reasonably place wireless sensor node i.e. can detect power operating state, save
Wiring link has been gone, substantial amounts of cost and effort is saved.
Radio sensing network node has that finite energy, computing capability are limited, had a very wide distribution, network dynamic performance it is strong with
And in network the features such as data volume is big, determine network node operating system should meet small size of code, modularization, low-power consumption,
The requirement such as concurrent operations and robustness, this is that traditional operating system can not be met, such as μ COS-II, Vx-Works.This
Specific implementation is tried hard to minimum hardware from the embedded OS TinyOS for aiming at the customization of wireless built sensing network
Support the concurrent intensive of network sensor.TinyOS employs the non-prerequisite variable FCFS deprived in task scheduling
(First Come First Served) scheduling strategy, a task once obtains the CPU rights to use would not be by except interrupting it
Other outer tasks are interrupted.So when setting up task, just without distributing a stack space for each task, all appoints
Business shares a stack space, has saved the memory headroom of operating system, and also saves when task context switches switching
Time.
2. software flow
In Design of System Software, the communication mechanism between radio sensing network node is emphasis, how reasonable design node
Between transceiving data mechanism be whole design have to solve key issue.Software function mainly include data acquisition and
Denoising, the implementation of routing algorithm and it is wirelessly transferred.Now illustrate software flow by taking radio communication as an example, see Figure 10.Sensing
Device network uses broadcast communication mode, and each node is allocated a unique ID, when node receives a packet, first
The ID of the packet header is taken out compared with the ID of oneself, if unanimously, receiving data, otherwise abandoning.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (3)
1. the electrical energy power quality disturbance recognition methods based on on-line training weighed SVM, it is characterised in that comprise the following steps：
S1, the different characteristics for common Power Quality Disturbance, set up the mathematical modeling of disturbing signal；
S2, the Generalized Morphological Filters based on mathematical morphology construction ADAPTIVE LMS ALGORITHM, are carried out to Power Quality Disturbance
Denoising；
S3, the Energy distribution by wavelet transformation extraction Power Quality Disturbance different frequency range；Electricity is extracted according to fractal theory
Can quality disturbance signal fractal dimension, the quantization spy that Energy distribution and fractal dimension collectively form disturbing signal leads and levies index；
S4, the time occurred according to electrical energy power quality disturbance assign noisy data different weights, realize on-line training and detection, right
Classification is identified in electrical energy power quality disturbance.
2. the electrical energy power quality disturbance recognition methods as claimed in claim 1 based on on-line training weighed SVM, it is characterised in that
The mathematical modeling of the disturbing signal of the step S1 includes
1. voltage dip (Voltage sag)
Wherein, 0.1 ＜ a ＜ 0.9 are temporary range of decrease degree, T ＜ t2-t1＜ 8T are temporarily drop duration, t1For moment, t takes place2For
Finish time, T is the cycle；
2. voltage swell (Voltage swell)
Wherein, 1.1<a<1.8 be temporary increasing degree degree, T<t2-t1<8T is temporarily liter duration, t1For moment, t takes place2To terminate
Moment；
3. voltage interruption (Interruption)
Wherein, 0<a<0.1 is interruption amplitude, T<t2-t1<8T is duration of interruption, t1For moment, t takes place2For at the end of
Carve；
4. transient state (Oscillatory transients) is vibrated
Wherein, 0.1 ＜ a ＜ 0.8 are vibration amplitude peak, 5<λ<15 be vibration frequency relative coefficient, 10<p<100 decline for vibration
Subtract coefficient, T ＜ t2-t1＜ 8T are duration of oscillation, t1For moment, t takes place2For finish time；
5. harmonic wave (Harmonics)
Wherein, 0.05 ＜ a3The ＜ a of ＜ 1,0.055The ＜ a of ＜ 1,0.057＜ 1 is 3 times, 5 times and 7 subharmonic amplitudes, for other times
Harmonic content is typically smaller, not impact analysis result；
6. voltage pulsation (Voltage fluctuation)
Wherein, UmFor work frequency carrier voltage magnitude, aUmCos λ ω t are frequency-modulated wave sinusoidal voltage waveform, and ω is work frequency carrier voltage
Angular frequency, 0.1 ＜ a ＜ 0.5,0.1<λ<0.5.
3. the electrical energy power quality disturbance recognition methods as claimed in claim 1 based on on-line training weighed SVM, it is characterised in that
The Generalized Morphological Filters of the ADAPTIVE LMS ALGORITHM are：
It is One-dimension Time Series to define input signal x (n), and g (n) is structural element, and structural element g (n) length is less than signal x
(n) length.For two structural element g1 (n) and g2 (n),And g1 (n) length is less than g2 (n) length
Degree, then constitute Generalized Morphological open-close, close-opening operation；
Wherein, o represents opening operation, represents closed operation.The output expression formula of so Generalized Morphological Filters is as follows：
Y (n)=α y1(x)+βy2(x) (9)
In formula：α and β are weight coefficients, and alpha+beta=1
Assuming that noise-free signal is Q, then least mean-square error is：
<mrow>
<mi>e</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<mi>Q</mi>
<mo>-</mo>
<mi>Y</mi>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>=</mo>
<mi>e</mi>
<mrow>
<mo>(</mo>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>Q</mi>
<mo>-</mo>
<mi>Y</mi>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>10</mn>
<mo>)</mo>
</mrow>
<mo>.</mo>
</mrow>
2
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710328990.5A CN107123987A (en) | 2017-05-11 | 2017-05-11 | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710328990.5A CN107123987A (en) | 2017-05-11 | 2017-05-11 | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107123987A true CN107123987A (en) | 2017-09-01 |
Family
ID=59727564
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710328990.5A Pending CN107123987A (en) | 2017-05-11 | 2017-05-11 | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107123987A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108053095A (en) * | 2017-11-22 | 2018-05-18 | 全球能源互联网研究院有限公司 | A kind of electrical energy power quality disturbance affair character extracting method and system |
| CN108052863A (en) * | 2017-11-10 | 2018-05-18 | 华北电力大学（保定） | Electrical energy power quality disturbance recognition methods based on the maximum variance method of development |
| CN108562811A (en) * | 2018-03-12 | 2018-09-21 | 西安理工大学 | Complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term |
| CN109061354A (en) * | 2018-09-10 | 2018-12-21 | 华南理工大学 | A kind of electrical energy power quality disturbance recognition methods based on improvement PSO and SVM |
| CN109145762A (en) * | 2018-07-27 | 2019-01-04 | 西南石油大学 | A kind of cable accessory Recognition of Partial Discharge based on mathematical morphology and fractal theory |
| CN109188220A (en) * | 2018-08-31 | 2019-01-11 | 中国电力科学研究院有限公司 | A kind of method and system of time domain index that assessing transmission line of electricity corona current signal |
| CN109324247A (en) * | 2018-11-13 | 2019-02-12 | 广东电网有限责任公司 | A kind of Power Quality Disturbance characteristic vector pickup method |
| CN109738686A (en) * | 2019-03-14 | 2019-05-10 | 广东电网有限责任公司 | A kind of electrical power system transient overvoltage quick discrimination method and device |
| CN109787222A (en) * | 2019-01-04 | 2019-05-21 | 三峡大学 | Distribution based on support vector machines is transported to electric system transient stability discrimination method |
| CN110048430A (en) * | 2019-05-08 | 2019-07-23 | 国网福建省电力有限公司莆田供电公司 | A kind of voltage-sensitive consumer networks weak spot recognition methods |
| CN110110762A (en) * | 2019-04-18 | 2019-08-09 | 东南大学 | Source discrimination method temporarily drops based on PDT-SVM in a kind of meter and lightning stroke situation |
| CN110824282A (en) * | 2019-11-20 | 2020-02-21 | 湖南铁路科技职业技术学院 | High-voltage charging pile electric energy quality monitoring system |
| CN110989363A (en) * | 2019-12-27 | 2020-04-10 | 广东电网有限责任公司电力科学研究院 | Electric energy quality control method and device based on deep learning |
| CN110989020A (en) * | 2019-12-12 | 2020-04-10 | 核工业北京地质研究院 | Filtering method and system for noise interference of audio magnetotelluric data |
| CN112232386A (en) * | 2020-09-27 | 2021-01-15 | 国网福建省电力有限公司莆田供电公司 | Voltage sag severity prediction method based on support vector machine |
| CN113702767A (en) * | 2021-08-31 | 2021-11-26 | 中南大学 | Island direct-current microgrid fault diagnosis method based on wavelet sliding window energy |
| CN113884969A (en) * | 2021-09-24 | 2022-01-04 | 广西电网有限责任公司电力科学研究院 | Error threshold value determination method for power quality monitoring device by utilizing fractal dimension detection |
| CN116207864A (en) * | 2023-04-28 | 2023-06-02 | 佰聆数据股份有限公司 | Method and system for controlling power equipment in low-voltage area based on Internet of things |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447464A (en) * | 2015-11-23 | 2016-03-30 | 广东工业大学 | Electric energy quality disturbance recognition and classification method based on PSO |
-
2017
- 2017-05-11 CN CN201710328990.5A patent/CN107123987A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447464A (en) * | 2015-11-23 | 2016-03-30 | 广东工业大学 | Electric energy quality disturbance recognition and classification method based on PSO |
Non-Patent Citations (1)
| Title |
|---|
| 石佳等: "《一种基于数学形态学的扰动信号分形检测方法》", 《电力系统及其自动化学报》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052863A (en) * | 2017-11-10 | 2018-05-18 | 华北电力大学（保定） | Electrical energy power quality disturbance recognition methods based on the maximum variance method of development |
| CN108053095A (en) * | 2017-11-22 | 2018-05-18 | 全球能源互联网研究院有限公司 | A kind of electrical energy power quality disturbance affair character extracting method and system |
| CN108053095B (en) * | 2017-11-22 | 2024-03-19 | 全球能源互联网研究院有限公司 | Power quality disturbance event feature extraction method and system |
| CN108562811A (en) * | 2018-03-12 | 2018-09-21 | 西安理工大学 | Complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term |
| CN108562811B (en) * | 2018-03-12 | 2020-05-22 | 西安理工大学 | Bidirectional long-short term memory-based complex power quality disturbance analysis method |
| CN109145762A (en) * | 2018-07-27 | 2019-01-04 | 西南石油大学 | A kind of cable accessory Recognition of Partial Discharge based on mathematical morphology and fractal theory |
| CN109145762B (en) * | 2018-07-27 | 2022-02-01 | 西南石油大学 | Cable accessory partial discharge identification method based on mathematical morphology and fractal theory |
| CN109188220A (en) * | 2018-08-31 | 2019-01-11 | 中国电力科学研究院有限公司 | A kind of method and system of time domain index that assessing transmission line of electricity corona current signal |
| CN109061354A (en) * | 2018-09-10 | 2018-12-21 | 华南理工大学 | A kind of electrical energy power quality disturbance recognition methods based on improvement PSO and SVM |
| CN109324247A (en) * | 2018-11-13 | 2019-02-12 | 广东电网有限责任公司 | A kind of Power Quality Disturbance characteristic vector pickup method |
| CN109324247B (en) * | 2018-11-13 | 2020-02-04 | 广东电网有限责任公司 | Electric energy quality disturbance signal feature vector extraction method |
| CN109787222A (en) * | 2019-01-04 | 2019-05-21 | 三峡大学 | Distribution based on support vector machines is transported to electric system transient stability discrimination method |
| CN109738686A (en) * | 2019-03-14 | 2019-05-10 | 广东电网有限责任公司 | A kind of electrical power system transient overvoltage quick discrimination method and device |
| CN109738686B (en) * | 2019-03-14 | 2020-11-10 | 广东电网有限责任公司 | Method and device for quickly judging transient overvoltage of power system |
| CN110110762A (en) * | 2019-04-18 | 2019-08-09 | 东南大学 | Source discrimination method temporarily drops based on PDT-SVM in a kind of meter and lightning stroke situation |
| CN110048430A (en) * | 2019-05-08 | 2019-07-23 | 国网福建省电力有限公司莆田供电公司 | A kind of voltage-sensitive consumer networks weak spot recognition methods |
| CN110824282A (en) * | 2019-11-20 | 2020-02-21 | 湖南铁路科技职业技术学院 | High-voltage charging pile electric energy quality monitoring system |
| CN110989020A (en) * | 2019-12-12 | 2020-04-10 | 核工业北京地质研究院 | Filtering method and system for noise interference of audio magnetotelluric data |
| CN110989363B (en) * | 2019-12-27 | 2022-01-25 | 广东电网有限责任公司电力科学研究院 | Electric energy quality control method and device based on deep learning |
| CN110989363A (en) * | 2019-12-27 | 2020-04-10 | 广东电网有限责任公司电力科学研究院 | Electric energy quality control method and device based on deep learning |
| CN112232386A (en) * | 2020-09-27 | 2021-01-15 | 国网福建省电力有限公司莆田供电公司 | Voltage sag severity prediction method based on support vector machine |
| CN112232386B (en) * | 2020-09-27 | 2024-03-15 | 国网福建省电力有限公司莆田供电公司 | Voltage sag severity prediction method based on support vector machine |
| CN113702767A (en) * | 2021-08-31 | 2021-11-26 | 中南大学 | Island direct-current microgrid fault diagnosis method based on wavelet sliding window energy |
| CN113884969A (en) * | 2021-09-24 | 2022-01-04 | 广西电网有限责任公司电力科学研究院 | Error threshold value determination method for power quality monitoring device by utilizing fractal dimension detection |
| CN113884969B (en) * | 2021-09-24 | 2023-07-18 | 广西电网有限责任公司电力科学研究院 | Error threshold determining method for detecting power quality monitoring device by using fractal dimension |
| CN116207864A (en) * | 2023-04-28 | 2023-06-02 | 佰聆数据股份有限公司 | Method and system for controlling power equipment in low-voltage area based on Internet of things |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107123987A (en) | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM | |
| Nguyen et al. | Power quality disturbance classification utilizing S-transform and binary feature matrix method | |
| Luo et al. | Robust optimal operation of active distribution network based on minimum confidence interval of distributed energy beta distribution | |
| CN101441231B (en) | Harmonic electric energy metering error analytical apparatus | |
| CN201218822Y (en) | Electric energy quality on-line monitoring device | |
| CN108169553A (en) | Electric energy gauging method under the conditions of distorted signal based on wavelet transformation and curve matching | |
| Cui et al. | HVDC transmission line fault localization base on RBF neural network with wavelet packet decomposition | |
| CN111308260A (en) | Electric energy quality monitoring and electric appliance fault analysis system based on wavelet neural network and working method thereof | |
| Nguyen et al. | Wavelet-based Neural Network for recognition of faults at NHABE power substation of the Vietnam power system | |
| CN111091223B (en) | Matching short-term load prediction method based on intelligent sensing technology of Internet of things | |
| CN103475247A (en) | Active and reactive dispatchable control device of distributed photovoltaic power station | |
| CN103219724B (en) | A kind of intelligent grid multi objective control system and method | |
| Yao et al. | A hybrid model to forecast wind speed based on wavelet and neural network | |
| Joga et al. | Identification of harmonic sources in smart grid using systematic feature extraction from non-active powers | |
| CN104699903A (en) | Intelligent substation secondary signal transmission system evaluation apparatus and method | |
| CN109713680A (en) | Based on the area power grid voltage quickly regulating method for improving Sensitivity Method | |
| Zhu et al. | Research on electrical power quality disturbance recognition method based on edge computing and LightGBM | |
| CN108880303A (en) | A kind of rectifier model predictive control method introducing big signal stabilization constraint | |
| Zhang et al. | Optimization of Flue Gas Desulphurization Technologies Based on Cloud Model and Kernel Vector Space Model | |
| Li et al. | Applying Hybrid MCDM Methods for Power Transmission System Evaluation and Selection | |
| Yang et al. | Research on Fault Diagnosis of Spark Discharge in Transformer Based on Sound+ Improved BP Neural Network | |
| Hong et al. | Establishment and application of the CW-MEEM model for post-evaluation of power grid construction projects | |
| CN109309384A (en) | A kind of power quality adjusting method and electric energy regulator | |
| CN207380147U (en) | A kind of electric energy quality on-line monitoring device | |
| CN206378528U (en) | A kind of intelligent telemetering type voltage quality monitoring analyzer |
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: 20170901 |
|
| RJ01 | Rejection of invention patent application after publication |