CN106874950A - A kind of method for identifying and classifying of transient power quality recorder data - Google Patents
A kind of method for identifying and classifying of transient power quality recorder data Download PDFInfo
- Publication number
- CN106874950A CN106874950A CN201710076722.9A CN201710076722A CN106874950A CN 106874950 A CN106874950 A CN 106874950A CN 201710076722 A CN201710076722 A CN 201710076722A CN 106874950 A CN106874950 A CN 106874950A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- quality
- classification
- time
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The invention discloses a kind of method for identifying and classifying of transient power quality recorder data, comprise the following steps:Quality of power supply recorder data is obtained, and training data and test data are screened in data;Module time-frequency matrixes are transformed the data into, and extracts the characteristic vector of module time-frequency matrixes;According to characteristic vector, the grader of the quality of power supply recorder data based on BP neural network is set up;Set up training sample and test sample;Training sample is input into grader, using the BP neural network of PSO algorithm optimization graders, the Optimum Classification device of the quality of power supply recorder data based on PSO BP neural networks is obtained;Testing feature vector is input into Optimum Classification device, and receives the testing classification of Optimum Classification device output;Judge testing classification and expect whether testing classification is consistent;When testing classification is consistent with testing classification is expected, testing classification is exported.The method for identifying and classifying of the transient power quality recorder data has the advantages that recognition efficiency is high, recognition accuracy is high and strong antijamming capability.
Description
Technical field
The present invention relates to power quality analysis technical field, more particularly to a kind of identification of transient power quality recorder data
Sorting technique.
Background technology
With science and technology and the high speed development of national economy, the new energy such as photovoltaic and wind-powered electricity generation is extensive simultaneously in power system
Net, additionally, increasing Large Copacity nonlinear-load extensive use in power system, such as electric railway, metallurgical smelting
Deng so that the power quality problem of power system is on the rise.To grasp the influence that power quality problem is caused to production activity,
The grid-connected place of many nonlinear-loads be provided with electric energy quality monitoring point be used to realize to the grid entry point quality of power supply not between
Disconnected monitoring.Equipment for monitoring power quality in addition to the stationary power quality data under the normal operating conditions that can draw stable state,
Transient power quality recorder data that can also be in the case of recording exceptional.Accurately and efficiently quality of power supply recorder data is divided
Analysis, identification and classification, help to find and solve the problems, such as power system in time, it is ensured that power system safety and stability runs.
At present, the identification and classification to quality of power supply recorder data mainly includes two steps, the i.e. extraction of data characteristics
With the identification of disturbance type.With the method for identifying and classifying stream of the transient power quality recorder data of the prior art shown in Fig. 1
Journey schematic diagram, with based on wavelet transformation and based on PSO (Particle Swarm Optimization, particle group optimizing in Fig. 1
Algorithm) BP (Back Propagation, backpropagation) neutral net quality of power supply recorder data sorting technique as a example by, the party
Method is to carry out multi-resolution decomposition to recorder data by using wavelet transformation, obtains the energy feature of recorder data on each yardstick,
And the individual features that will be extracted are input into PSO-BP neutral nets, realize that the disturbance type of recorder data is classified.
But, it is above-mentioned based in wavelet transformation and PSO-BP neutral net quality of power supply recorder data sorting techniques, first
Recorder data feature is extracted using wavelet transformation, then with the recorder data feature extracted directly as identification model input
Amount, carries out the training study of neutral net.Lack because wavelet transformation has spectral leakage, easily affected by noise and transformation results
The problems such as intuitive, and input quantity data are complicated and huge, cause neural metwork training to recognize that difficulty is big, training speed is slow, instruction
Practice accuracy low, finally influence identification and the classifying quality of whole method.
The content of the invention
To overcome problem present in correlation technique, the present invention to provide a kind of identification of transient power quality recorder data point
Class method.
A kind of method for identifying and classifying of the transient power quality recorder data for providing according to embodiments of the present invention, including it is following
Step:Quality of power supply recorder data is obtained, and training data and test data are screened in data;Transform the data into mould time-frequency
Matrix, and extract the characteristic vector of module time-frequency matrixes;According to characteristic vector, the quality of power supply record ripple based on BP neural network is set up
The grader of data;Training sample and test sample are set up, wherein, training feature vector and phase of the training sample by training data
Prestige training classification composition, test sample is made up of the testing feature vector and expectation testing classification of test data;By training sample
Input grader, using the BP neural network of PSO algorithm optimization graders, obtains the quality of power supply based on PSO-BP neutral nets
The Optimum Classification device of recorder data;Testing feature vector is input into Optimum Classification device, and receives the test of Optimum Classification device output
Classification;Judge testing classification and expect whether testing classification is consistent;When testing classification is consistent with testing classification is expected, output is surveyed
Examination classification.
Alternatively, the method for identifying and classifying of transient power quality recorder data, also includes:When testing classification is tested with expectation
When classifying inconsistent, using test sample as error sample;Error sample is incorporated to training sample and is concentrated as training sample
A training sample.
Alternatively, quality of power supply recorder data is obtained, and training data and test data is screened in data, including:Obtain
Take real-time quality of power supply recorder data;By randomly choosing training data in real-time quality of power supply recorder data;By real-time electric energy matter
Amount recorder data is used as test data.
Alternatively, obtain real-time quality of power supply recorder data and obtain history quality of power supply recorder data in database;By
Training data is randomly choosed in real-time quality of power supply recorder data and/or history quality of power supply recorder data;By real-time electric energy matter
Amount recorder data is used as test data.
Alternatively, quality of power supply recorder data is obtained, and training data and test data is screened in data, including:Base
In MATLAB software emulations generation emulation data, wherein, emulation data include voltage dip data, voltage swell data, voltage
Interrupt data, transient oscillation data and transient state pulse data;Obtain emulation data and real-time quality of power supply recorder data;Will emulation
Data are used as training data;Using real-time quality of power supply recorder data as test data.
Alternatively, module time-frequency matrixes are transformed the data into, and extracts characteristic vector, including:Transform the data into mould time-frequency
Matrix, obtains initial characteristicses vector;The coefficient correlation of initial characteristicses vector is calculated, initial characteristicses vector is screened, obtained
Characteristic vector.
Alternatively, module time-frequency matrixes are transformed the data into, and extracts characteristic vector, including:When being carried out at equal intervals to data
Between sample, obtain discrete time data;Discrete time data are transformed into multiple time-frequency matrix;By each element in multiple time-frequency matrix
Modulus computing is carried out, module time-frequency matrixes are obtained;According to module time-frequency matrixes, time-frequency characteristic curve P is built, wherein, time-frequency characteristic is bent
Line includes time-frequency contour P1, time amplitude envelope curve P2, frequency amplitude envelope curve P3 and frequency standard difference curve P4;Root
According to time-frequency characteristic curve, data characteristics is extracted with construction feature vector F, wherein, characteristic vector F=[F1, F2, F3, F4, F5,
F6, F7, F8], wherein, F1 is the mean square deviation of highest frequency contour amplitude in P1;F2 is highest frequency contour amplitude in P1
Average value;F3 is the average energy in P2;F4 is the amplitude factor in P2;F5 is the time width corresponding to octuple fundamental frequency
Value curve mean square deviation;F6 is that the time amplitude curve extreme value corresponding to octuple fundamental frequency is poor;F7 is the mean square deviation of P3;F8 is P4
Subduplicate average value.
Alternatively, using PSO algorithms and the BP neural network of BP algorithm Optimum Classification device, including:First pass through PSO algorithms pair
The input layer of BP neural network is to hidden layer connection weight, hidden layer to output layer connection weight, hidden layer threshold value and output layer
Threshold value carries out optimizing assignment;Again using BP algorithm to the input layer of optimizing assignment to hidden layer connection weight, hidden layer to exporting
Layer connection weight, hidden layer threshold value and output layer threshold value are trained.
Alternatively, training sample and test sample are set up, including:The training feature vector of training data is obtained, and, survey
Try the testing feature vector of data;Screening is instructed with the classification of training data matching degree highest data as expectation in database
Practice classification, and, the classification with test data matching degree highest data is used as test training classification;By training feature vector and
Expect to train classification composition training sample, and, testing feature vector and expectation testing classification constitute test sample.
Alternatively, training sample and test sample are set up, including:Obtain the training feature vector of training data;Will training
The corresponding emulation data classification of data is used as expectation training classification;By training feature vector and expectation training classification composition training sample
This;Obtain the testing feature vector of test data;Screening and the classification of test data matching degree highest data in database
As test training classification;By testing feature vector and expectation testing classification composition test sample.
The technical scheme that embodiments of the invention are provided can include the following benefits:
A kind of method for identifying and classifying of transient power quality recorder data provided in an embodiment of the present invention, including following step
Suddenly:Quality of power supply recorder data is obtained, and training data and test data are screened in data;Transform the data into mould time-frequency square
Battle array, and extract the characteristic vector of module time-frequency matrixes;According to characteristic vector, the quality of power supply record wave number based on BP neural network is set up
According to grader;Training sample and test sample are set up, wherein, training feature vector and expectation of the training sample by training data
Training classification composition, test sample is made up of the testing feature vector and expectation testing classification of test data;Training sample is defeated
Enter grader, using the BP neural network of PSO algorithm optimization graders, obtain the quality of power supply based on PSO-BP neutral nets and record
The Optimum Classification device of wave number evidence;Testing feature vector is input into Optimum Classification device, and receives the test point of Optimum Classification device output
Class;Judge testing classification and expect whether testing classification is consistent;When testing classification is consistent with testing classification is expected, output test
Classification.The method carries out feature extraction to recorder data using the reversible S-transformation analysis method of time-frequency, both many with wavelet transformation
The ability of variability analysis, avoids the select permeability of Short Time Fourier Transform window function again, and various features can be used in advance to record ripple
Disturbance of data is recognized, is difficult simultaneously by noise jamming with stronger intuitive.Therefore, the transient power quality recorder data
Method for identifying and classifying has the advantages that recognition efficiency is high, recognition accuracy is high and strong antijamming capability.
It should be appreciated that the general description of the above and detailed description hereinafter are only exemplary and explanatory, not
Can the limitation present invention.
Brief description of the drawings
Accompanying drawing herein is merged in specification and constitutes the part of this specification, shows and meets implementation of the invention
Example, and be used to explain principle of the invention together with specification.
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
The accompanying drawing to be used needed for having technology description is briefly described, it should be apparent that, for those of ordinary skill in the art
Speech, without having to pay creative labor, can also obtain other accompanying drawings according to these accompanying drawings.
A kind of flow of the method for identifying and classifying of transient power quality recorder data that Fig. 1 is provided for prior art is illustrated
Figure;
Fig. 2 is that a kind of flow of the method for identifying and classifying of transient power quality recorder data provided in an embodiment of the present invention is shown
It is intended to;
Fig. 3 is the schematic flow sheet of the step of embodiment of the present invention one is provided S110;
Fig. 4 is the schematic flow sheet of the step of embodiment of the present invention two is provided S110;
Fig. 5 is the schematic flow sheet of the step of embodiment of the present invention three is provided S110;
Fig. 6 is the schematic flow sheet of the step of embodiment of the present invention one is provided S140;
Fig. 7 is the schematic flow sheet of the step of embodiment of the present invention three is provided S140.
Specific embodiment
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Following description is related to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment
Described in implementation method do not represent and the consistent all implementation methods of the present invention.Conversely, they be only with it is such as appended
The example of the consistent apparatus and method of some aspects being described in detail in claims, of the invention.
It is an object of the invention to overcome the deficiencies in the prior art, there is provided a kind of scientific and reasonable, recognition efficiency and accuracy rate
Height, the method for identifying and classifying of the transient power quality recorder data of strong antijamming capability, Fig. 2 is to show according to embodiments of the present invention
A kind of transient power quality recorder data method for identifying and classifying schematic flow sheet, specifically, the present invention provides following real
Example is applied to be illustrated details.
Embodiment one
The method for identifying and classifying of the transient power quality recorder data that the present invention is provided, comprises the following steps:
Step S110, obtains quality of power supply recorder data x (t), and screening training data and test number in data x (t)
According to.Specifically, as shown in figure 3, step S110 is further comprising the steps of:
Step S111, Real-time Collection simultaneously obtains the real-time quality of power supply recorder data of user side or system side, real-time electric energy
Quality recorder data is voltage magnitude change curve not in the same time, as quality of power supply recorder data x (t).
Step S112, the stochastical sampling in real-time quality of power supply recorder data, selection is instructed to follow-up BP neural network
Experienced training data.Step S113, using real-time quality of power supply recorder data as test data, to the quality of power supply situation of user
Detected.
Step S120, transforms the data into module time-frequency matrixes, and extract the characteristic vector of module time-frequency matrixes.Specifically, walk
Rapid S120's is further comprising the steps of:
Step S121, the treatment of time sampling at equal intervals is carried out to quality of power supply recorder data x (t), obtains discrete time number
According to x (kT), wherein k=0,1,2 ..., N-1, T be the sampling interval, N is sampling sum.
Discrete time data x (kT) are carried out discrete S-transformation by step S122, obtain multiple time-frequency matrix S:
In formula,It is the multiple time-frequency matrix S after conversion with S [jT, 0], the row correspondence frequency in multiple time-frequency matrix S
Rate, row correspondence time, i.e., the amplitude at behavioral data x (t) of time-frequency matrix S a certain moment, in the distribution of different frequency, is classified as again
The amplitude of a certain frequency of data x (t) is in distribution not in the same time.
Step S123, modulus computing is carried out by each element in multiple time-frequency matrix S, obtains module time-frequency matrixes STA, mould time-frequency
Matrix STABehavioral data x (t) a certain moment amplitude in the distribution of different frequency, be classified as the amplitude of a certain frequency of data x (t)
In distribution not in the same time.Module time-frequency matrixes STAIntroduce width and frequency be inversely proportional the Gaussian window of conversion, with frequency phase
The resolution ratio of pass, both with the ability of poly-tropic rate analysis, can avoid the need for selection window type and width and window width again
The not enough problem such as fixed, can shift to an earlier date various features and be recognized for disturbance of data, and with stronger intuitive and be difficult
It is affected by noise.
Step S124, according to module time-frequency matrixes STA, time-frequency characteristic curve P is built, wherein, time-frequency characteristic curve includes time-frequency
Contour P1, time amplitude envelope curve P2, frequency amplitude envelope curve P3 and frequency standard difference curve P4, in each curve, when
Between represented using sampled point, frequency is represented using normalized frequency, f=n/NT.
Step S125, according to time-frequency characteristic curve, extracts data characteristics with construction feature vector F, wherein, characteristic vector F=
[F1, F2, F3, F4, F5, F6, F7, F8], wherein, F1 is the mean square deviation of highest frequency contour amplitude in P1;F2 is highest in P1
The average value of frequency contour amplitude;F3 is the average energy in P2In formula, xiIt is ith sample point in P2
Corresponding amplitude, N is total number of sample points;F4 is the amplitude factor in P2
In formula, FNIt is maxima and minima sum in the P2 of standard sine data;F5 is the time width corresponding to octuple fundamental frequency
Value curve mean square deviation;F6 is that the time amplitude curve extreme value corresponding to octuple fundamental frequency is poor;F7 is the mean square deviation of P3;F8 is P4
Subduplicate average value.
To improve the recognition efficiency and accuracy rate of later stage grader, step S120 is further comprising the steps of:
Step S126, according to step S121 to S125, transforms the data into module time-frequency matrixes, obtain 8 initial characteristicses to
Amount.
Step S127, calculate initial characteristicses vector coefficient correlation, initial characteristicses vector is screened, obtain feature to
Amount.By calculating 8 spearman coefficient correlations (product moment correlation coefficient) of initial characteristicses vector, filter out special with noisy data
The larger combination of eigenvectors of correlation is levied, characteristic vector F is formedbest, as the input information of later stage Classification and Identification device.
Spearman coefficient correlation computing formula are as follows:
In formula, d is respectively to feature XiEach pair observation (x after order is taken with data Yi, the difference of order y), n is that observation is right
Number.
Step S130, according to characteristic vector, using characteristic vector as input vector, using classification results as output vector,
Set up the grader of the quality of power supply recorder data based on BP neural network.Specifically, S130 is comprised the following steps:
Step S131, determines input layer number n, node in hidden layer l, output layer nodes h.In the present invention, n
Value is characterized vectorial F or FbestDimension, be improve grader classification accuracy and classification effectiveness, the preferred F of value of nbest
Dimension;The value of h is the class number of data, including five kinds of common noisy datas (voltage dip data, voltage swell numbers
According to, voltage interruption data, transient oscillation data and transient state pulse data);The value of l meets l=log2h。
Step S132, chooses the transmission function of hidden layer and output layer.
Step S140, sets up the training sample and test specimens of the grader of the electrical energy power quality disturbance based on BP neural network
This, wherein, training sample is made up of the training feature vector and expectation training classification of training data, and test sample is by test data
Testing feature vector and expect testing classification composition.Specifically, as shown in fig. 6, step S140 is comprised the following steps:
Step S141, obtains the training feature vector of training data, and, the testing feature vector of test data.
Step S142, using the historical data stored in database and corresponding history classification results, in database
Screening is classified with the classification of training data matching degree highest data as expectation training, and, with test data matching degree most
The classification of data high can be matched as test training classification, specific matching process using Waveform Matching or matrix element.
Step S143, by training feature vector and expect training classification composition training sample, and, testing feature vector and
Expect that testing classification constitutes test sample.
Step S150, grader is input into by training sample, using the BP neural network of PSO algorithm optimization graders, is obtained
The Optimum Classification device of the quality of power supply recorder data based on PSO-BP neutral nets, specifically, step S150 also includes following step
Suddenly:
Step S151, is arrived to the input layer of the BP neural network by PSO algorithms to hidden layer connection weight, hidden layer
Output layer connection weight, hidden layer threshold value and output layer threshold value carry out optimizing assignment.
Step S152, using BP algorithm to the input layer of optimizing assignment to hidden layer connection weight, hidden layer to output layer
Connection weight, hidden layer threshold value and output layer threshold value are trained.By PSO to identification disaggregated model in connection weight and
Threshold value is carried out after preliminary optimizing assignment, using BP algorithm according to these connection weights and the further optimizing of threshold value, so as to obtain
The optimal value of connection weight and threshold value in identification disaggregated model.
Step S160, is input into Optimum Classification device, and receive the testing classification of Optimum Classification device output by testing feature vector.
Step S170, is the accuracy for further improving grader, using the phase obtained by database in step S140
Testing classification is hoped, testing classification is judged and is expected whether testing classification is consistent.
Step S180, when testing classification is consistent with testing classification is expected, reaches the accuracy standard required by test, defeated
Go out testing classification.According to the quantity of output testing classification, the classification performance to grader is evaluated and tested, for example, the test of input
The quantity of data is 100, and the testing classification that can be accurately identified and export is 98, and at this moment the classification accuracy of grader can
It is recorded as 98%.
Step S190, when testing classification is with expecting that testing classification is inconsistent, then follows the steps below:
Step S191, when testing classification is with expecting that testing classification is inconsistent, using this test sample as wrong sample
This.Error sample may be considered the test sample for not meeting data historian data characteristics.
Why step S192, grader cannot obtain the testing classification knot matched with historical data feature in database
Really, because lacking the classification capacity to the such data characteristics of error sample, therefore in order to improve the accurate of grader
Property, error sample is incorporated to training sample and the training sample concentrated as training sample.
Embodiment two
Step S110 is in addition to the implementation method that embodiment one is provided, as shown in figure 4, following step can also be specifically included
Suddenly:
The real-time quality of power supply of step S114, Real-time Collection and the quality of power supply for obtaining user side or system side records wave number
According to, at the same can with history quality of power supply recorder data stored in acquisition database, real-time quality of power supply recorder data and
History quality of power supply recorder data is amplitude change curve not in the same time, is quality of power supply recorder data.
Step S115, the stochastical sampling in real-time quality of power supply recorder data, selection is instructed to follow-up BP neural network
Experienced training data;Or in real-time quality of power supply recorder data and the history quality of power supply recorder data stochastical sampling, select it is right
The training data that follow-up BP neural network is trained;Or in the history quality of power supply recorder data stochastical sampling, select to rear
The training data that continuous BP neural network is trained.
Step S116, using real-time quality of power supply recorder data as test data, the quality of power supply situation to user is carried out
Detection.
Embodiment three
Step S110 is in addition to the implementation method that embodiment one and embodiment two are provided, as shown in figure 5, can also specifically wrap
Include following steps:
Step S117, based on MATLAB software emulations generation emulation data, wherein, the emulation data include voltage dip
Data, voltage swell data, voltage interruption data, transient oscillation data and transient state pulse data.Specifically, the quality of power supply is imitative
The expression formula of true data is as shown in table 1.
The expression formula of the emulation data of the quality of power supply of table 1
Step S118, obtains emulation data and real-time quality of power supply recorder data.
Step S119, using emulation data as training data, using real-time quality of power supply recorder data as test data.
Now, it is, it is known that so need not be by data historian data characteristics to emulation number due to emulating the type of data
Judgement classification is carried out according to type, therefore, as shown in fig. 7, step S140 specifically includes following steps:
Step S144, obtains the training feature vector of training data.
Step S145, using the corresponding emulation data classification of training data as expectation training classification.
Step S146, by training feature vector and expectation training classification composition training sample.
Step S147, obtains the testing feature vector of test data.
Step S148, screening divides with the classification of test data matching degree highest data as test training in database
Class.
Step S149, by testing feature vector and expectation testing classification composition test sample.
In this specification between each embodiment identical similar part mutually referring to.Especially for embodiment two
For embodiment three, because it is substantially similar to embodiment one, so description is fairly simple, related part is referring to method reality
Apply the explanation in example.Invention described above implementation method is not intended to limit the scope of the present invention..
The method for identifying and classifying of transient power quality recorder data provided in an embodiment of the present invention is become using the reversible S of time-frequency
Change analysis method carries out feature extraction to quality of power supply recorder data, both with the ability of wavelet transformation poly-tropic rate analysis, keeps away again
Exempt from the select permeability of Short Time Fourier Transform window function, can in advance to various features be used for recorder data disturbed depth, with compared with
Strong intuitive is difficult by noise jamming simultaneously.Therefore, the method for identifying and classifying of the transient power quality recorder data has and knows
Other efficiency high, recognition accuracy be high and advantage of strong antijamming capability.
Those skilled in the art considering specification and practice here after disclosure of the invention, will readily occur to it is of the invention its
Its embodiment.The application is intended to any modification of the invention, purposes or adaptations, these modifications, purposes or
Person's adaptations follow general principle of the invention and including undocumented common knowledge in the art of the invention
Or conventional techniques.Description and embodiments are considered only as exemplary, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be appreciated that the invention is not limited in the precision architecture being described above and be shown in the drawings, and
And can without departing from the scope carry out various modifications and changes.The scope of the present invention is only limited by appended claim.
Claims (10)
1. a kind of method for identifying and classifying of transient power quality recorder data, it is characterised in that comprise the following steps:
Quality of power supply recorder data is obtained, and training data and test data are screened in the data;
The data are transformed into module time-frequency matrixes, and extract the characteristic vector of the module time-frequency matrixes;
According to the characteristic vector, the grader of the quality of power supply recorder data based on BP neural network is set up;
Training sample and test sample are set up, wherein, training feature vector and phase of the training sample by the training data
Prestige training classification composition, the test sample is made up of the testing feature vector and expectation testing classification of the test data;
The training sample is input into the grader, using the BP neural network of grader described in PSO algorithm optimizations, base is obtained
In the Optimum Classification device of the quality of power supply recorder data of PSO-BP neutral nets;
The testing feature vector is input into the Optimum Classification device, and receives the testing classification of the Optimum Classification device output;
Judge whether the testing classification is consistent with the expectation testing classification;
When the testing classification is consistent with the expectation testing classification, testing classification is exported.
2. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that it is described temporarily
The method for identifying and classifying of state quality of power supply recorder data, also includes:
When the testing classification is inconsistent with the expectation testing classification, using the test sample as error sample;
The error sample is incorporated to the training sample and the training sample concentrated as training sample.
3. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that described to obtain
Quality of power supply recorder data is taken, and training data and test data are screened in the data, including:
Obtain real-time quality of power supply recorder data;
By randomly choosing training data in the real-time quality of power supply recorder data;
Using the real-time quality of power supply recorder data as test data.
4. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that described to obtain
Quality of power supply recorder data is taken, and training data and test data are screened in the data, including:
Obtain real-time quality of power supply recorder data and obtain history quality of power supply recorder data in database;
By randomly choosing training number in the real-time quality of power supply recorder data and/or the history quality of power supply recorder data
According to;
Using the real-time quality of power supply recorder data as test data.
5. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that described to obtain
Quality of power supply recorder data is taken, and training data and test data are screened in the data, including:
Based on MATLAB software emulations generation emulation data, wherein, the emulation data include voltage dip data, voltage swell
Data, voltage interruption data, transient oscillation data and transient state pulse data;
Obtain emulation data and the real-time quality of power supply recorder data;
Using the emulation data as training data;
Using the real-time quality of power supply recorder data as test data.
6. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that it is described will
The data are transformed into module time-frequency matrixes, and extract characteristic vector, including:
The data are transformed into module time-frequency matrixes, initial characteristicses vector is obtained;
The coefficient correlation of the initial characteristicses vector is calculated, the initial characteristicses vector is screened, obtain characteristic vector.
7. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that it is described will
The data are transformed into module time-frequency matrixes, and extract characteristic vector, including:
Time sampling at equal intervals is carried out to the data, discrete time data are obtained;
The discrete time data are transformed into multiple time-frequency matrix;
Each element in the multiple time-frequency matrix is carried out into modulus computing, module time-frequency matrixes are obtained;
According to the module time-frequency matrixes, time-frequency characteristic curve P is built, wherein, the time-frequency characteristic curve P includes time-frequency contour
P1, time amplitude envelope curve P2, frequency amplitude envelope curve P3 and frequency standard difference curve P4;
According to the time-frequency characteristic curve P, data characteristics is extracted with construction feature vector F, wherein, the characteristic vector F=
[F1, F2, F3, F4, F5, F6, F7, F8], wherein, F1 is the mean square deviation of highest frequency contour amplitude in P1;F2 be P1 in most
The average value of high-frequency contour amplitude;F3 is the average energy in P2;F4 is the amplitude factor in P2;F5 be octuple fundamental wave frequently
Time amplitude curve mean square deviation corresponding to rate;F6 is that the time amplitude curve extreme value corresponding to octuple fundamental frequency is poor;F7 is
The mean square deviation of P3;F8 is the subduplicate average value of P4.
8. the method for identifying and classifying of transient power quality recorder data according to claim 1, it is characterised in that described to adopt
With the BP neural network of grader described in PSO algorithm optimizations, including:
PSO algorithms are first passed through to connect the input layer of the BP neural network to hidden layer connection weight, hidden layer to output layer
Weights, hidden layer threshold value and output layer threshold value carry out optimizing assignment;
Again using BP algorithm to the input layer of optimizing assignment to hidden layer connection weight, hidden layer to output layer connection weight, hidden
Threshold value containing layer and output layer threshold value are trained.
9. the method for identifying and classifying of the transient power quality recorder data according to any one of claim 3 or 4, its feature exists
In, it is described to set up training sample and test sample, including:
The training feature vector of the training data is obtained, and, the testing feature vector of the test data;
The classification with the training data matching degree highest data is screened in database as expectation training classification, and,
Classification with the test data matching degree highest data is used as test training classification;
By the training feature vector and expect training classification composition training sample, and, the testing feature vector and expectation
Testing classification constitutes test sample.
10. the method for identifying and classifying of transient power quality recorder data according to claim 5, it is characterised in that described
Training sample and test sample are set up, including:
Obtain the training feature vector of the training data;
Using the corresponding emulation data classification of the training data as expectation training classification;
By the training feature vector and expectation training classification composition training sample;
Obtain the testing feature vector of the test data;
The classification with the test data matching degree highest data is screened in database as test training classification;
By the testing feature vector and expectation testing classification composition test sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710076722.9A CN106874950A (en) | 2017-02-13 | 2017-02-13 | A kind of method for identifying and classifying of transient power quality recorder data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710076722.9A CN106874950A (en) | 2017-02-13 | 2017-02-13 | A kind of method for identifying and classifying of transient power quality recorder data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106874950A true CN106874950A (en) | 2017-06-20 |
Family
ID=59165930
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710076722.9A Pending CN106874950A (en) | 2017-02-13 | 2017-02-13 | A kind of method for identifying and classifying of transient power quality recorder data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106874950A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107681375A (en) * | 2017-10-25 | 2018-02-09 | 济南大学 | A kind of automatic shutoff energy-saving socket |
CN107769197A (en) * | 2017-11-14 | 2018-03-06 | 国网江苏省电力公司电力科学研究院 | A kind of voltage sag source alignment system based on grid equipment topology |
CN108805204A (en) * | 2018-06-12 | 2018-11-13 | 东北大学 | Electrical energy power quality disturbance analytical equipment based on deep neural network and its application method |
CN108846411A (en) * | 2018-05-03 | 2018-11-20 | 武汉理工大学 | Main reducing gear method for diagnosing faults based on pairing sparse Bayesian extreme learning machine |
CN108919003A (en) * | 2018-07-06 | 2018-11-30 | 华北电力大学 | A kind of test of sensitive equipment voltage dip resistance characteristics and data processing method |
CN110889396A (en) * | 2019-12-12 | 2020-03-17 | 国家电网有限公司大数据中心 | Energy internet disturbance classification method and device, electronic equipment and storage medium |
CN112529066A (en) * | 2020-12-07 | 2021-03-19 | 南京工程学院 | Voltage sag source identification method based on BAS-BP classifier model |
CN113033837A (en) * | 2021-03-05 | 2021-06-25 | 国网电力科学研究院武汉南瑞有限责任公司 | Artificial intelligence fault identification system and method based on transient waveform of power transmission line |
CN113361427A (en) * | 2021-06-11 | 2021-09-07 | 温州大学 | WT and GA-PNN based composite power quality disturbance classification method |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102831433A (en) * | 2012-06-06 | 2012-12-19 | 西南交通大学 | Method for classifying electric energy quality mixing disturbances based on multi-feature quantity of time-frequency domain |
JP2013074695A (en) * | 2011-09-27 | 2013-04-22 | Meiji Univ | Device, method and program for predicting photovoltaic generation |
CN103679209A (en) * | 2013-11-29 | 2014-03-26 | 广东领域安防有限公司 | Sparse theory based character recognition method |
CN104134068A (en) * | 2014-08-12 | 2014-11-05 | 江苏理工学院 | Monitored vehicle characteristic representation and classification method based on sparse coding |
CN105469144A (en) * | 2015-11-19 | 2016-04-06 | 东北大学 | Mobile communication user loss prediction method based on particle classification and BP neural network |
CN106052854A (en) * | 2016-06-13 | 2016-10-26 | 浙江理工大学 | Grinding machine grinding chatter fault on-line diagnosis method |
CN106250904A (en) * | 2016-05-18 | 2016-12-21 | 国网新疆电力公司电力科学研究院 | Based on Power Disturbance analyser and the sorting technique of improving S-transformation |
CN106295802A (en) * | 2016-08-01 | 2017-01-04 | 安徽农业大学 | A kind of Folium Camelliae sinensis based on particle cluster algorithm Optimized BP Neural Network storage method chronological classification |
-
2017
- 2017-02-13 CN CN201710076722.9A patent/CN106874950A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013074695A (en) * | 2011-09-27 | 2013-04-22 | Meiji Univ | Device, method and program for predicting photovoltaic generation |
CN102831433A (en) * | 2012-06-06 | 2012-12-19 | 西南交通大学 | Method for classifying electric energy quality mixing disturbances based on multi-feature quantity of time-frequency domain |
CN103679209A (en) * | 2013-11-29 | 2014-03-26 | 广东领域安防有限公司 | Sparse theory based character recognition method |
CN104134068A (en) * | 2014-08-12 | 2014-11-05 | 江苏理工学院 | Monitored vehicle characteristic representation and classification method based on sparse coding |
CN105469144A (en) * | 2015-11-19 | 2016-04-06 | 东北大学 | Mobile communication user loss prediction method based on particle classification and BP neural network |
CN106250904A (en) * | 2016-05-18 | 2016-12-21 | 国网新疆电力公司电力科学研究院 | Based on Power Disturbance analyser and the sorting technique of improving S-transformation |
CN106052854A (en) * | 2016-06-13 | 2016-10-26 | 浙江理工大学 | Grinding machine grinding chatter fault on-line diagnosis method |
CN106295802A (en) * | 2016-08-01 | 2017-01-04 | 安徽农业大学 | A kind of Folium Camelliae sinensis based on particle cluster algorithm Optimized BP Neural Network storage method chronological classification |
Non-Patent Citations (5)
Title |
---|
MISHRA S 等: "Detection and Classification of Power Quality Disturbances Using S-Transform and Probabilistic Neural Network", 《IEEE TRANSACTIONS ON POWER DELIVERY》 * |
WANG X 等: "Hybrid PSO-BP Based Probabilistic Neural Network for Power Transformer Fault Diagnosis", 《 SECOND INTERNATIONAL SYMPOSIUM ON INTELLIGENT INFORMATION TECHNOLOGY APPLICATION》 * |
卢绪祥 等: "基于EMD及灰色关联度的滑动轴承润滑状态故障诊断研究", 《动力工程学报》 * |
李晓 等: "基于小波变换与PSO-BP神经网络的电能质量扰动分类", 《高等函授学报(自然科学版)》 * |
赵强强: "基于S变换和神经网络的电能质量扰动检测分类", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107681375A (en) * | 2017-10-25 | 2018-02-09 | 济南大学 | A kind of automatic shutoff energy-saving socket |
CN107769197A (en) * | 2017-11-14 | 2018-03-06 | 国网江苏省电力公司电力科学研究院 | A kind of voltage sag source alignment system based on grid equipment topology |
CN108846411A (en) * | 2018-05-03 | 2018-11-20 | 武汉理工大学 | Main reducing gear method for diagnosing faults based on pairing sparse Bayesian extreme learning machine |
CN108805204A (en) * | 2018-06-12 | 2018-11-13 | 东北大学 | Electrical energy power quality disturbance analytical equipment based on deep neural network and its application method |
CN108919003A (en) * | 2018-07-06 | 2018-11-30 | 华北电力大学 | A kind of test of sensitive equipment voltage dip resistance characteristics and data processing method |
CN108919003B (en) * | 2018-07-06 | 2020-10-23 | 华北电力大学 | Sensitive equipment voltage sag tolerance characteristic testing and data processing method |
CN110889396A (en) * | 2019-12-12 | 2020-03-17 | 国家电网有限公司大数据中心 | Energy internet disturbance classification method and device, electronic equipment and storage medium |
CN112529066A (en) * | 2020-12-07 | 2021-03-19 | 南京工程学院 | Voltage sag source identification method based on BAS-BP classifier model |
CN112529066B (en) * | 2020-12-07 | 2024-04-16 | 南京工程学院 | Voltage sag source identification method based on BAS-BP classifier model |
CN113033837A (en) * | 2021-03-05 | 2021-06-25 | 国网电力科学研究院武汉南瑞有限责任公司 | Artificial intelligence fault identification system and method based on transient waveform of power transmission line |
CN113361427A (en) * | 2021-06-11 | 2021-09-07 | 温州大学 | WT and GA-PNN based composite power quality disturbance classification method |
CN113361427B (en) * | 2021-06-11 | 2024-03-15 | 温州大学 | Composite electric energy quality disturbance classification method based on WT and GA-PNN |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106874950A (en) | A kind of method for identifying and classifying of transient power quality recorder data | |
CN109614981B (en) | Power system intelligent fault detection method and system based on spearman level-dependent convolutional neural network | |
CN110108992B (en) | Cable partial discharge fault identification method and system based on improved random forest algorithm | |
CN103941254A (en) | Soil physical property classification recognition method and device based on geological radar | |
CN111562612B (en) | Deep learning microseismic event identification method and system based on attention mechanism | |
CN103487832A (en) | Method for classifying supervised waveforms in three-dimensional seismic signal | |
CN108964046B (en) | Short-time disturbed trajectory-based power system transient stability evaluation method | |
CN104766090B (en) | A kind of Coherent Noise in GPR Record method for visualizing based on BEMD and SOFM | |
CN108135003A (en) | The construction method and system of interference type identification model | |
CN111628494B (en) | Low-voltage distribution network topology identification method and system based on logistic regression method | |
CN104966161A (en) | Electric energy quality recording data calculating analysis method based on Gaussian mixture model | |
CN110166154A (en) | A kind of software radio spectrum monitoring knowledge method for distinguishing neural network based | |
Sivakumar et al. | Dynamics of monthly rainfall-runoff process at the Gota basin: A search for chaos | |
CN113687433B (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
CN102364490B (en) | Automatic synchronization recognition method based on hierarchical analyzing model | |
CN112327104A (en) | Fault detection and positioning method for power distribution network with distributed power supply | |
CN114152980A (en) | Method and device for rapidly and automatically producing seismic source mechanism solution | |
CN114169249A (en) | Artificial intelligence identification method for high-resistance grounding fault of power distribution network | |
CN113889198A (en) | Transformer fault diagnosis method and equipment based on oil chromatogram time-frequency domain information and residual error attention network | |
CN104834816A (en) | Short-term wind speed prediction method | |
CN115310499B (en) | Industrial equipment fault diagnosis system and method based on data fusion | |
CN109919101A (en) | A kind of user experience evaluation method and system based on cell phone client | |
CN109815108A (en) | A kind of combined test set of uses case priorization sort method and system based on weight | |
CN104516858A (en) | Phase diagram matrix method for nonlinear dynamic behavior analysis | |
CN117556369B (en) | Power theft detection method and system for dynamically generated residual error graph convolution neural network |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170620 |