CN108562811A - Complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term - Google Patents
Complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term Download PDFInfo
- Publication number
- CN108562811A CN108562811A CN201810198781.8A CN201810198781A CN108562811A CN 108562811 A CN108562811 A CN 108562811A CN 201810198781 A CN201810198781 A CN 201810198781A CN 108562811 A CN108562811 A CN 108562811A
- Authority
- CN
- China
- Prior art keywords
- long term
- pqd
- way shot
- layer
- sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Complex Calculations (AREA)
- Measurement Of Current Or Voltage (AREA)
Abstract
The invention discloses the complicated electrical energy power quality disturbance analysis methods remembered based on two-way shot and long term, specially：Acquire total sample of voltage or several or complicated PQD using the basic PQD of 7 classes of mathematical model acquisition and its various combination composition of current signal in electric system to be detected；Sample marks and is converted into sequence form, and then sample is divided into training set and test set；It builds two-way shot and long term Memory Neural Networks model and is trained；Then over-fitting judgement is carried out, hyper parameter is adjusted if there is over-fitting, then re -training, is so recycled, until not occurring over-fitting；PQD judgements are carried out using trained neural network model again, input data is signal sequence data, and output data is the electric energy type corresponding to each data in sequence.The present invention solves the disadvantages that recognition accuracy existing in the prior art is low, realization process is complicated, real-time is poor, can not carry out essence to disturbance start/stop time and problem is accurately positioned.
Description
Technical field
The invention belongs to power quality analysis and detection method technical field, it is related to a kind of based on two-way shot and long term remembering
Complicated electrical energy power quality disturbance analysis method.
Background technology
Accurately identifying for disturbance type is complicated electrical energy power quality disturbance (Power Quality Disturbance, PQD) point
The premise and basis of analysis.Electrical energy power quality disturbance (Power Quality Disturbance, PQD) can be divided into basic PQD and answer
Miscellaneous PQD.According to the time response of disturbance, basic PQD be divided into as stable state disturbance (mainly include harmonic wave/m-Acetyl chlorophosphonazo, fluctuation etc.) and
Transient disturbance (mainly including temporarily drop, temporary liter, interruption, oscillation transient state, impulse transients etc.).Complicated PQD is disturbed by a variety of differences
Dynamic type, different strength of turbulence, different start/stop times basic PQD be combined, refer in particular to be superimposed disturbing for transient state component
It is dynamic.
Currently, the research about electrical energy power quality disturbance type identification problem is mainly using the side learnt based on shallow-layer
Method can be attributed to two links of Characteristic Extraction and pattern-recognition, i.e., by after original signal is converted and is reconstructed from
Middle extraction disturbing signal characteristic quantity, and then disturbance type identification is carried out using shallow Models such as neural network or support vector machines,
There are following common problems for such methods：
(1) majority takes into consideration only basic PQD or two kinds of basic PQD (majority is all harmonic wave and other basic PQD) are constituted
Complicated PQD, but with the increase of basic PQD types included in complicated PQD, the coupling interaction phenomenon between characteristic quantity is serious, leads
Recognition accuracy is caused significantly to decline or can not solve the problems, such as at all the disturbance type identification of complicated PQD；
(2) there is good effect to stable state disturbance, but to transient disturbance, extracts their amplitude and frequecy characteristic merely
They cannot be reflected well, cause accuracy rate low；
(3) what feature selects, how selected characteristic must deeply understand characteristics of signals or according to expert of the art's
Abundant engineering experience goes to attempt, and results in manual perturbation features extraction and selection course is complex.In addition to this, confine
In the natural mode of " pattern-recognition after first feature extraction ", treat relationship between the two with isolating, recognition accuracy seriously according to
Rely the complexity and redundancy that disturbance type recognition procedure is increased in the perturbation features amount manually designed in advance, it is difficult to ensure
The real-time of identification process, is not suitable for application on site.
Invention content
The object of the present invention is to provide a kind of complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term, solutions
Recognition accuracy existing in the prior art of having determined is low, realization process is complicated, real-time is poor, can not be carried out to disturbance start/stop time
The shortcomings that essence is accurately positioned problem.
The technical solution adopted in the present invention is the complicated electrical energy power quality disturbance analysis side based on the memory of two-way shot and long term
Method is specifically implemented according to the following steps：
Step 1, voltage is acquired in electric system to be detected using measuring instrument or current signal is several or use formula
(1) mathematical model shown in, obtain it is a series of comprising temporarily rise, temporarily drop, interrupt, oscillation transient state, impulse transients, harmonic wave/humorous
The total sample for the complicated PQD that wave, the basic PQD of 7 classes of fluctuation and its various combination are constituted；
Complicated PQD wherein comprising temporarily liter, temporary drop, interruption, oscillation transient state, impulse transients, harmonic wave/m-Acetyl chlorophosphonazo, fluctuation believes
The unified parameters neutralizing analysis mathematical modeling formula (1) of number s (t) is shown below：
Wherein, δ (t), A, f0(t)、θ1(t)、N、an(t)、fn(t)、ψn(t) disturbance related with amplitude is represented：δ(t)
For time-varying function, A, f0(t)、θ1(t) be respectively fundamental signal amplitude, frequency, phase, N, an(t)、fn(t)、ψn(t) respectively
Represent the fluctuation number of envelope component, the amplitude of n times envelope signal, frequency, phase；H、h、bh(t)、θh(t) it represents and fundamental wave frequency
The related integral frequency harmonizing wave disturbance of rate：H、h、bh(t)、θh(t) number of harmonic component, overtone order, h subharmonic are respectively represented
Amplitude, phase；K、ck(t)、fk(t)、Represent the m-Acetyl chlorophosphonazo disturbance unrelated with fundamental frequency：K、ck(t)、fk(t)、The number of m-Acetyl chlorophosphonazo component, amplitude, frequency, the phase of k m-Acetyl chlorophosphonazo are respectively represented；M、αm、βm、dm、fm、
τmRepresent transient disturbance：M、αm、βm、dm、fm、τmIt respectively represented oscillation transient state component number, vibrated transient state m times
Initial time, end time, amplitude, frequency, phase, decay factor；μ (t) is noise component(s)；
Step 2, sample marks
It will be defined and mark respectively according to type is disturbed included in signal through the collected PQD samples of step 1
Note；
Step 3, the PQD samples marked through step 2 are converted to sequence form；
Step 4, the sample after step 3 serializing is divided into training set and test set, wherein training set accounts for total sample
70%, test set data account for the 30% of total sample；
Step 5, build two-way shot and long term Memory Neural Networks model, including from down to up sequentially connected importation,
Implicit layer segment and output layer segment；
Step 6, the two-way shot and long term Memory Neural Networks model built through step 5 is trained, every time training time
Each training data in training set is gone through, traversal is referred to as a generation every time, and neural network model is made to carry out multiple generations
Training passes through several from generation to generation, obtains trained two-way shot and long term Memory Neural Networks model；
Step 7, over-fitting judges
Using the two-way shot and long term Memory Neural Networks model after step 6 trains, with 80% data in test set
It is tested, obtains data accuracy, then tested with the test set data of residue 20%, if accuracy declines to a great extent
Then there is over-fitting, then adjusts the hyper parameter of two-way shot and long term Memory Neural Networks model, then the two-way length of re -training
Short-term memory neural network model executes over-fitting judgement, so recycles again after training, until do not occur over-fitting, to
Obtain the good two-way shot and long term Memory Neural Networks model of generalization；
Step 8, using trained two-way shot and long term neural network model carries out PQD judgements again through step 7, number is inputted
According to for signal sequence data, output data is the electric energy type corresponding to each data in sequence.
The features of the present invention also characterized in that
Complicated PQD in step 1 includes all combining forms that two kinds or more basic PQD are constituted.
Following principle should be followed when being combined to complicated PQD using formula (1) in step 1：A, same parameters cannot be simultaneously
Two separate modes of mutation occurs；B, different parameters can mutate simultaneously；C, the presence of additive disturbance is not become by parameter
The limitation of change.
Sequence in step 3 includes two parts, and first part is signal sequence, and the sampling interval sets according to the sampling time
Depending on setting；Second part is sequence label, and sequence label marks out the type belonging to each element, sequence label and sampling sequence
Element in row corresponds.
Input layer segment in step 5 only includes input layer, and it includes multilayer hidden layer to imply layer segment, is contained in hidden layer
There is two-way shot and long term memory layer, full articulamentum, abandon layer, output layer segment is Soft-Max layers.
Input layer input data format is [samples, sequence-length, dim], wherein samples is input
Data sample amount, identical as sequence number, sequence-length is list entries length, and dim is data dimension.
Remaining neural net layer is connect by activation primitive with preceding layer neural net layer in addition to first layer input layer.
The beneficial effects of the invention are as follows
(1) present invention can autonomous learning disturbs directly from original bottom data characteristic information, can be from simple
The complicated implicit features for being difficult to quantify are extracted in explicit features, complicated PQD signals are more comprehensively described, maximum journey
The integrality that ensure that complicated PQD information on degree avoids the cumbersome manual characteristic extraction procedure in traditional shallow-layer study, improves
Rapidity and accuracy rate, the disturbance start/stop time positioning accuracy of disturbance type identification.
(2) it is not necessarily to make any hypothesis or processing to noise in advance, avoids introducing unnecessary error, there is the stronger anti-back of the body
Scape noise jamming ability, in addition to this it is possible to realize the organic unity of feature extraction and pattern-recognition, the two is carried out at the same time simultaneously
It is generated in training simultaneously, end-to-end online real-time processing may be implemented, effectively reduce algorithm complexity.
(3) two-way shot and long term Memory Neural Networks can not only refer to the member before currentElement during judgement
Element can also refer to the element after currentElement, and realization is bi-directionally referenced, so as to solve traditional shot and long term Memory Neural Networks
In judgement problem of dtmf distortion DTMF, obtain accurate each Basic power quality disturbances start/stop time.
Description of the drawings
Fig. 1 is a kind of embodiment for the complicated electrical energy power quality disturbance analysis method remembered the present invention is based on two-way shot and long term
Figure；
Fig. 2 is the disturbance for the 0.4-2.6s that the simulation electric energy quality monitoring terminal of Noise in the embodiment of the present invention detects
Wave data figure；
Fig. 3 is the disturbance for the 3.0-5.4s that the simulation electric energy quality monitoring terminal of Noise in the embodiment of the present invention detects
Wave data figure.
Specific implementation mode
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
The present invention is based on the complicated electrical energy power quality disturbance analysis methods of two-way shot and long term memory, specifically real according to the following steps
It applies：
Step 1, voltage is acquired in electric system to be detected using measuring instrument or current signal is several or use formula
(1) mathematical model shown in, obtain it is a series of comprising temporarily rise, temporarily drop, interrupt, oscillation transient state, impulse transients, harmonic wave/humorous
Wave, fluctuation the basic PQD of 7 classes and its various combination constitute complicated PQD total sample, wherein complexity PQD comprising two kinds and with
All combining forms that upper basic PQD is constituted, and when being combined to complicated PQD, following principle should be followed：A, same parameters cannot
Two separate modes of mutation occurs simultaneously；B, different parameters can mutate simultaneously；C, the presence of additive disturbance is not joined
The limitation of number variation；
Complicated PQD wherein comprising temporarily liter, temporary drop, interruption, oscillation transient state, impulse transients, harmonic wave/m-Acetyl chlorophosphonazo, fluctuation believes
The unified parameters neutralizing analysis mathematical modeling formula (1) of number s (t) is shown below：
Wherein, δ (t), A, f0(t)、θ1(t)、N、an(t)、fn(t)、ψn(t) disturbance related with amplitude is represented：δ(t)
For time-varying function, A, f0(t)、θ1(t) be respectively fundamental signal amplitude, frequency, phase, N, an(t)、fn(t)、ψn(t) respectively
Represent the fluctuation number of envelope component, the amplitude of n times envelope signal, frequency, phase；H、h、bh(t)、θh(t) it represents and fundamental wave frequency
The related integral frequency harmonizing wave disturbance of rate：H、h、bh(t)、θh(t) number of harmonic component, overtone order, h subharmonic are respectively represented
Amplitude, phase；K、ck(t)、fk(t)、Represent the m-Acetyl chlorophosphonazo disturbance unrelated with fundamental frequency：K、ck(t)、fk(t)、The number of m-Acetyl chlorophosphonazo component, amplitude, frequency, the phase of k m-Acetyl chlorophosphonazo are respectively represented；M、αm、βm、dm、fm、
τmRepresent transient disturbance：M、αm、βm、dm、fm、τmIt respectively represented oscillation transient state component number, vibrated transient state m times
Initial time, end time, amplitude, frequency, phase, decay factor；μ (t) is noise component(s)；
Step 2, sample marks
It will be defined and mark respectively according to type is disturbed included in signal through the collected PQD samples of step 1
Note；
Step 3, the PQD samples marked through step 2 are converted to sequence form, wherein sequence include two parts, first
A part is signal sequence, depending on the sampling interval is arranged according to the sampling time；Second part is sequence label, sequence label mark
Go out the type belonging to each element, sequence label is corresponded with the element in sample sequence；
Step 4, the sample after step 3 serializing is divided into training set and test set, wherein training set accounts for total sample
70%, test set data account for the 30% of total sample；
Step 5, build two-way shot and long term Memory Neural Networks model, including from down to up sequentially connected importation,
Implicit layer segment and output layer segment, wherein input layer segment only includes input layer, and it includes that multilayer is implicit to imply layer segment
Layer, containing two-way shot and long term memory layer, full articulamentum, discarding layer in hidden layer, it is Soft-Max layers to export layer segment；Input layer
Input data format is [samples, sequence-length, dim], wherein samples is input sample of data amount, with sequence
Row number is identical, and sequence-length is list entries length, and dim is data dimension；Remaining god in addition to first layer input layer
It is connect with preceding layer neural net layer by activation primitive through network layer.
Step 6, the two-way shot and long term Memory Neural Networks model built through step 5 is trained, every time training time
Each training data in training set is gone through, traversal is referred to as a generation every time, and neural network model is made to carry out multiple generations
Training passes through several from generation to generation, obtains trained two-way shot and long term Memory Neural Networks model；
Step 7, over-fitting judges
Using the two-way shot and long term Memory Neural Networks model after step 6 trains, with 80% data in test set
It is tested, obtains data accuracy, then tested with the test set data of residue 20%, if accuracy declines to a great extent
Then there is over-fitting, then adjusts the hyper parameter of two-way shot and long term Memory Neural Networks model, then the two-way length of re -training
Short-term memory neural network model executes over-fitting judgement, so recycles again after training, until do not occur over-fitting, to
Obtain the good two-way shot and long term Memory Neural Networks model of generalization；
Step 8, using trained two-way shot and long term neural network model carries out PQD judgements again through step 7, number is inputted
According to for signal sequence data, output data is the electric energy type corresponding to each data in sequence.
Embodiment
Based on the complicated electrical energy power quality disturbance analysis method of two-way shot and long term memory, it is specifically implemented according to the following steps：
Step 1, voltage in electric system to be detected is acquired using measuring instrument or current signal is several or use is public
Mathematical model shown in formula (1), obtain it is a series of comprising temporarily rise, temporarily drop, interrupt, oscillation transient state, impulse transients, harmonic wave/
Harmonic wave, fluctuation the basic PQD of 7 classes and its various combination constitute complicated PQD total sample, wherein complicated PQD include two kinds and
All combining forms that the above basic PQD is constituted, and in step 1 using formula (1) when being combined to complicated PQD, should follow with
Lower principle：A, two separate modes of mutation cannot occur simultaneously for same parameters；B, different parameters can mutate simultaneously；c、
The presence of additive disturbance is not limited by Parameters variation；
Complicated PQD wherein comprising temporarily liter, temporary drop, interruption, oscillation transient state, impulse transients, harmonic wave/m-Acetyl chlorophosphonazo, fluctuation believes
The unified parameters neutralizing analysis mathematical modeling formula (1) of number s (t) is shown below：
Wherein, δ (t), A, f0(t)、θ1(t)、N、an(t)、fn(t)、ψn(t) disturbance related with amplitude is represented：δ(t)
For time-varying function, A, f0(t)、θ1(t) be respectively fundamental signal amplitude (V), frequency (Hz), phase (rad/s), N, an(t)、
fn(t)、ψn(t) number for fluctuating envelope component, the amplitude (V) of n times envelope signal, frequency (Hz), phase (rad/ are respectively represented
s)；H、h、bh(t)、θh(t) integral frequency harmonizing wave disturbance related with fundamental frequency is represented：H、h、bh(t)、θh(t) it respectively represents humorous
Number, overtone order, the amplitude (V) of h subharmonic, the phase (rad/s) of wave component；K、ck(t)、fk(t)、Representative and base
The unrelated m-Acetyl chlorophosphonazo disturbance of wave frequency rate：K、ck(t)、fk(t)、The number of m-Acetyl chlorophosphonazo component, k m-Acetyl chlorophosphonazo are respectively represented
Amplitude (V), frequency (Hz), phase (rad/s)；M、αm、βm、dm、fm、τmTransient disturbance is represented, for example, oscillation is temporary
State, impulse transients, different disturbance parameter value ranges can indicate different disturbance types：M、αm、βm、dm、fm、τm
Oscillation transient state component number, the initial time (s) for vibrating transient state m times, end time (s), amplitude (V), frequency are respectively represented
(Hz), phase (rad/s), decay factor (s)；μ (t) is noise component(s)；
Step 2, sample marks
It will be defined and mark respectively according to type is disturbed included in signal through the collected PQD samples of step 1
Note；
Step 3, the PQD samples marked through step 2 are converted to sequence form, wherein sequence include two parts, first
A part is signal sequence, depending on the sampling interval is arranged according to the sampling time；Second part is sequence label, sequence label mark
Go out the type belonging to each element, sequence label is corresponded with the element in sample sequence；In this example, PQD signals per second
Including 1000 sampled points, i.e., each signal sequence includes 1000 elements；Sequence label marks out belonging to each element
Type, sequence label element are corresponded with sample sequence element.
Step 4, the sample after step 3 serializing is divided into training set and test set, wherein training set accounts for total sample
70%, test set data account for the 30% of total sample；
Step 5, build two-way shot and long term Memory Neural Networks model, including from down to up sequentially connected importation,
Implicit layer segment and output layer segment, wherein input layer segment only includes input layer, and it includes that multilayer is implicit to imply layer segment
Layer, remember layer, full articulamentum containing two-way shot and long term in hidden layer, abandon layer (dropout layers), output layer segment is Soft-
Max layers；Pass through last layer of Soft-Max layers of sentencing come output sequence by the two-way shot and long term Memory Neural Networks after training
Disconnected result；Remaining neural net layer is connect by activation primitive with preceding layer neural net layer in addition to first layer input layer.Example
Such as：As shown in Figure 1, using one layer of input layer, using eight layers of hidden layer, hidden layer is followed successively by two-way shot and long term memory layer, abandons
Layer, two-way shot and long term remember layer, abandon layer, and two-way shot and long term remembers layer, abandon layer, and two-way shot and long term remembers layer, abandon layer, double
Remember layer, full articulamentum to shot and long term；It is output layer segment later, which is one layer Soft-Max layers；Input layer inputs number
It is [samples, sequence-length, dim] according to format, wherein samples is input sample of data amount, with sequence number
Identical, sequence-length is list entries length, and dim is data dimension；In this example, it is total that samples is set as sequence
Number, sequence-length 1000, dim are set as 1, i.e., only there are one dimensions for input data.Output sequence and list entries
Form is identical.Output content is type label corresponding with input signal sequence.In hidden layer, two-way shot and long term memory nerve
Network Element Layer is set as 500 implicit units, and two-way shot and long term memory layer n-steps sizes are sequence length and input data
Sequence-length sizes it is identical, be used herein as 1000.It is 0.3 to abandon layer loss ratio, and full articulamentum is 1000 lists
Member.Output layer is Soft-max layers, this layer is connected with the full articulamentum of last layer.Activation primitive uses Relu activation primitives.
Each layer of output data carries out standardization processing using batch standardization.
Step 6, the two-way shot and long term Memory Neural Networks model built through step 5 is trained, every time training time
Each training data in training set is gone through, traversal is referred to as a generation every time, and neural network model is made to carry out multiple generations
Training passes through several from generation to generation, obtains trained two-way shot and long term Memory Neural Networks model；In this example, the overall situation is used
The mode of random initializtion carries out parameter initialization, uses the training of Adam's optimizer or Momentum optimizers or SGD optimizations
Device, 5000 generations of training, learning rate 0.01, loss function are used as loss function using cross entropy or damage cross entropy
It loses function and is used as loss function using mean square deviation or mean difference.
Step 7, over-fitting judges
Using the two-way shot and long term Memory Neural Networks model after step 6 trains, with 80% data in test set
It is tested, obtains data accuracy, then tested with the test set data of residue 20%, if accuracy declines to a great extent
Then there is over-fitting, then adjusts the hyper parameter of two-way shot and long term Memory Neural Networks model, then the two-way length of re -training
Short-term memory neural network model executes over-fitting judgement, so recycles again after training, until do not occur over-fitting, to
Obtain the good two-way shot and long term Memory Neural Networks model of generalization；There is over-fitting and then adjust hyper parameter, as modification is lost
A layer loss ratio is abandoned, connection layer number, modification learning rate, change training generation, adjustment entirely is changed and implies layer number and (such as reduce two-way
Shot and long term remembers layer),
Step 8, using trained two-way shot and long term neural network model carries out PQD judgements again through step 7, number is inputted
According to for signal sequence data, output data is the electric energy type corresponding to each data in sequence；Input data at this time
Sequence-length is changed to 1, i.e., input data is signal sequence data one by one, and output data is each number in sequence
According to corresponding electric energy type.
Implying layer number in the present invention in neural network model can modify, and increase or decrease two-way shot and long term memory
Layer abandons layer, full articulamentum and the full articulamentum element number of modification.
Remaining neural net layer passes through activation primitive and preceding layer neural network in addition to first layer input layer in the present invention
Layer connection, wherein activation primitive can be ReLU or Leaky Relu or Sigmoid or tanh as activation primitive.
Neural network hyper parameter can be adjusted according to actual conditions in the present invention, such as quantity, the sequence of training generation
Window sequence in n-steps sizes, learning rate, list entries length, list entries dimension and two-way shot and long term memory layer
Size.
Simulation comparison, which is given below, proves the detection result of proposition method of the present invention：
According to formula (1) institute representation model, each parameter in above formula is changed at random in range, to every a kind of disturbing signal
300 data samples are generated, wherein 200 are used as training sample, 100 are used as test sample, and it is 20dB to be superimposed signal-to-noise ratio
White Gaussian noise.
Table 1 show based on two-way length in short-term Memory Neural Networks 48 kinds of PQD signals respectively noiseless and superposition believe
It makes an uproar than the disturbance type self adaption recognition result under two kinds of operating modes of noise for 20dB；
The disturbance type identification result of complicated PQD under 1 noiseless of table and noisy acoustic environment
Table 1 shares 5 column datas, and what first row indicated is the recognition accuracy of 7 kinds of basic PQD, and what secondary series indicated is 2 kinds
The complicated PQD recognition accuracies that basic PQD is constituted, one of which disturbance come from basic PQD shown in row 1, another kind disturbance
The basic PQD shown in the row 2, and so on, what 3~row of row 5 indicated respectively is the complexity being made of 3~5 kinds of basic PQD
PQD disturbs type identification accuracy rate.Such as, the 5th row the first row overstriking mark data fluctuation (98,97) can be regarded as in table：Altogether by
The complicated PQD that 5 kinds of basic PQD (temporary drop, harmonic wave/m-Acetyl chlorophosphonazo, oscillation transient state, pulse, fluctuation) are constituted is under noiseless operating mode
Recognition accuracy is 98%, and the disturbance type identification accuracy rate in the case where signal-to-noise ratio is 20dB operating modes is 97%.
Table 1 is made a general survey of, can obtain and such as draw a conclusion：1) the complicated electrical energy power quality disturbance analysis method gram based on the memory of two-way shot and long term
Traditional shallow-layer study basic PQD types included in feature extraction and the deficiency in disturbance type identification, complicated PQD are taken
Number influences very little to disturbed depth accuracy rate, and identification is accurately also greatly improved, the complexity PQD under noise-free case
Disturbance type identification accuracy rate is above 99%.2) when superposition signal-to-noise ratio is the ambient noise of 20dB, complicated PQD disturbances type is known
Other accuracy rate is above 97%, therefore the disturbance kind identification method has strong anti-noise ability, without making an uproar to original signal
Sound Processing for removing.
Moreover, it is desirable to which it is emphasized that the two-way shot and long term neural network model after the completion of training also has knowledge
Not fireballing advantage, can meet the requirement of real-time of actual power quality online monitoring system, and this requirement based on
It is but often implacable in the complicated PQD disturbance kind identification methods of traditional shallow-layer study.
As Figure 2-3, it is the disturbance waveform data that detect of simulation electric energy quality monitoring terminal of Noise, using double
The start/stop time for the whole elementary events for being included to it to shot and long term neural network model positions, obtained testing result
As shown in table 2, it can be seen that the algorithm can be accurately detected the disturbance start/stop time of each elementary event, meet actual monitoring system
System requires.
Table 2 disturbs start/stop time testing result
Claims (7)
1. the complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term, which is characterized in that specifically according to following step
It is rapid to implement：
Step 1, voltage is acquired in electric system to be detected using measuring instrument or current signal is several or using formula (1)
Mathematical model, obtain it is a series of comprising temporarily rise, temporarily drop, interrupt, oscillation transient state, impulse transients, harmonic wave/m-Acetyl chlorophosphonazo, fluctuation
The basic PQD of 7 classes and its various combination constitute complicated PQD total sample；
Wherein include temporarily rise, temporarily drop, interrupt, oscillation transient state, impulse transients, harmonic wave/m-Acetyl chlorophosphonazo, fluctuation complicated PQD signals s
(t) unified parameters neutralizing analysis mathematical modeling formula (1) is shown below：
Wherein, δ (t), A, f0(t)、θ1(t)、N、an(t)、fn(t)、ψn(t) disturbance related with amplitude is represented：When δ (t) is
Varying function, A, f0(t)、θ1(t) be respectively fundamental signal amplitude, frequency, phase, N, an(t)、fn(t)、ψn(t) it respectively represents
Fluctuate the number of envelope component, the amplitude of n times envelope signal, frequency, phase；H、h、bh(t)、θh(t) representing has with fundamental frequency
The integral frequency harmonizing wave of pass disturbs：H、h、bh(t)、θh(t) number of harmonic component, the width of overtone order, h subharmonic are respectively represented
Value, phase；K、ck(t)、fk(t)、Represent the m-Acetyl chlorophosphonazo disturbance unrelated with fundamental frequency：K、ck(t)、fk(t)、Point
The number of m-Acetyl chlorophosphonazo component, amplitude, frequency, the phase of k m-Acetyl chlorophosphonazo are not represented；M、αm、βm、dm、fm、θm(t)、τmIt represents
Transient disturbance：M、αm、βm、dm、fm、θm(t)、τmWhen having respectively represented oscillation transient state component number, the starting for vibrating transient state m times
Quarter, end time, amplitude, frequency, phase, decay factor；μ (t) is noise component(s)；
Step 2, sample marks
It will be defined and mark respectively according to type is disturbed included in signal through the collected PQD samples of step 1；
Step 3, the PQD samples marked through step 2 are converted to sequence form；
Step 4, the sample after step 3 serializing is divided into training set and test set, wherein training set accounts for total sample 70%, surveys
Examination collection data account for the 30% of total sample；
Step 5, two-way shot and long term Memory Neural Networks model is built, including sequentially connected importation, implicit from down to up
Layer segment and output layer segment；
Step 6, the two-way shot and long term Memory Neural Networks model built through step 5 is trained, every time training traversal instruction
Practice each training data concentrated, traversal is referred to as a generation every time, and neural network model is made to carry out multiple generation training,
Pass through several from generation to generation, obtains trained two-way shot and long term Memory Neural Networks model；
Step 7, over-fitting judges
Using the two-way shot and long term Memory Neural Networks model after step 6 trains, with 80% data in test set come into
Row test, obtains data accuracy, is then tested with the test set data of residue 20%, is gone out if accuracy declines to a great extent
Existing over-fitting, then adjust the hyper parameter of two-way shot and long term Memory Neural Networks model, then the two-way shot and long term of re -training
Memory Neural Networks model executes over-fitting judgement, so recycles again after training, until not occurring over-fitting, to obtain
The good two-way shot and long term Memory Neural Networks model of generalization；
Step 8, using through step 7, trained two-way shot and long term neural network model carries out PQD judgements again, input data is
Signal sequence data, output data are the electric energy type corresponding to each data in sequence.
2. the complicated electrical energy power quality disturbance analysis method according to claim 1 based on the memory of two-way shot and long term, feature
It is, the complicated PQD described in step 1 includes all combining forms that two kinds or more basic PQD are constituted.
3. the complicated electrical energy power quality disturbance analysis method according to claim 1 based on the memory of two-way shot and long term, feature
It is, following principle should be followed when being combined to complicated PQD using formula (1) in the step 1：A, same parameters cannot be same
The two separate modes of mutation of Shi Fasheng；B, different parameters can mutate simultaneously；C, the presence of additive disturbance is not by parameter
The limitation of variation.
4. the complicated electrical energy power quality disturbance analysis method according to claim 1 based on the memory of two-way shot and long term, feature
It is, the sequence described in step 3 includes two parts, and first part is signal sequence, and the sampling interval is according to the sampling time
Depending on setting；Second part is sequence label, and sequence label marks out the type belonging to each element, sequence label and sampling
Element in sequence corresponds.
5. the complicated electrical energy power quality disturbance analysis method according to claim 1 based on the memory of two-way shot and long term, feature
It is, the input layer segment described in step 5 only includes input layer, and implicit layer segment includes the multilayer hidden layer being sequentially connected,
Containing two-way shot and long term memory layer, full articulamentum, discarding layer in hidden layer, it is Soft-Max layers to export layer segment.
6. the complicated electrical energy power quality disturbance analysis method according to claim 5 based on the memory of two-way shot and long term, feature
It is, the input layer input data format is [samples, sequence-length, dim], wherein samples is input
Data sample amount, identical as sequence number, sequence-length is list entries length, and dim is data dimension.
7. the complicated electrical energy power quality disturbance analysis method according to claim 5 based on the memory of two-way shot and long term, feature
It is, remaining neural net layer is connect by activation primitive with preceding layer neural net layer in addition to first layer input layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810198781.8A CN108562811B (en) | 2018-03-12 | 2018-03-12 | Bidirectional long-short term memory-based complex power quality disturbance analysis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810198781.8A CN108562811B (en) | 2018-03-12 | 2018-03-12 | Bidirectional long-short term memory-based complex power quality disturbance analysis method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108562811A true CN108562811A (en) | 2018-09-21 |
| CN108562811B CN108562811B (en) | 2020-05-22 |
Family
ID=63531608
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810198781.8A Active CN108562811B (en) | 2018-03-12 | 2018-03-12 | Bidirectional long-short term memory-based complex power quality disturbance analysis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108562811B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214107A (en) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | A kind of ship's navigation behavior on-line prediction method |
| CN109508811A (en) * | 2018-09-30 | 2019-03-22 | 中冶华天工程技术有限公司 | Parameter prediction method is discharged based on principal component analysis and the sewage treatment of shot and long term memory network |
| CN110222953A (en) * | 2018-12-29 | 2019-09-10 | 北京理工大学 | A kind of power quality hybrid perturbation analysis method based on deep learning |
| CN110728195A (en) * | 2019-09-18 | 2020-01-24 | 武汉大学 | Power quality disturbance detection method based on YOLO algorithm |
| CN110829417A (en) * | 2019-11-14 | 2020-02-21 | 电子科技大学 | Electric power system transient stability prediction method based on LSTM double-structure model |
| CN111781244A (en) * | 2020-07-15 | 2020-10-16 | 中国科学院自动化研究所 | Infrared thermal imaging type coating detection method based on long-term and short-term memory network |
| WO2021129668A1 (en) * | 2019-12-24 | 2021-07-01 | 华为技术有限公司 | Neural network training method and device |
| CN113536655A (en) * | 2021-04-07 | 2021-10-22 | 北京聚树核科技有限公司 | Artificial intelligent deviation rectifying method and device for heliostat, electronic equipment and storage medium |
| CN114992138A (en) * | 2022-06-13 | 2022-09-02 | 湘潭大学 | Method for diagnosing faults of water supply pump of industrial circulating water system based on GAN-CNN-BilSTM |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101034389A (en) * | 2007-03-19 | 2007-09-12 | 江西省电力科学研究院 | Electrical energy power quality disturbance automatic identification method and system based on information fusion |
| CN105205560A (en) * | 2015-09-25 | 2015-12-30 | 国网甘肃省电力公司电力科学研究院 | Photovoltaic power supply power prediction method based on positive and negative error variable weights |
| CN106778923A (en) * | 2017-03-23 | 2017-05-31 | 广东电网有限责任公司珠海供电局 | A kind of Power Quality Disturbance sorting technique and device |
| CN107123987A (en) * | 2017-05-11 | 2017-09-01 | 华东交通大学 | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM |
| CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
-
2018
- 2018-03-12 CN CN201810198781.8A patent/CN108562811B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101034389A (en) * | 2007-03-19 | 2007-09-12 | 江西省电力科学研究院 | Electrical energy power quality disturbance automatic identification method and system based on information fusion |
| CN105205560A (en) * | 2015-09-25 | 2015-12-30 | 国网甘肃省电力公司电力科学研究院 | Photovoltaic power supply power prediction method based on positive and negative error variable weights |
| CN106778923A (en) * | 2017-03-23 | 2017-05-31 | 广东电网有限责任公司珠海供电局 | A kind of Power Quality Disturbance sorting technique and device |
| CN107123987A (en) * | 2017-05-11 | 2017-09-01 | 华东交通大学 | Electrical energy power quality disturbance recognition methods based on on-line training weighed SVM |
| CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 金燕等: "基于改进型神经网络的电能质量扰动识别", 《电力系统及其自动化》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109214107A (en) * | 2018-09-26 | 2019-01-15 | 大连海事大学 | A kind of ship's navigation behavior on-line prediction method |
| CN109508811A (en) * | 2018-09-30 | 2019-03-22 | 中冶华天工程技术有限公司 | Parameter prediction method is discharged based on principal component analysis and the sewage treatment of shot and long term memory network |
| CN110222953A (en) * | 2018-12-29 | 2019-09-10 | 北京理工大学 | A kind of power quality hybrid perturbation analysis method based on deep learning |
| CN110728195A (en) * | 2019-09-18 | 2020-01-24 | 武汉大学 | Power quality disturbance detection method based on YOLO algorithm |
| CN110829417A (en) * | 2019-11-14 | 2020-02-21 | 电子科技大学 | Electric power system transient stability prediction method based on LSTM double-structure model |
| WO2021129668A1 (en) * | 2019-12-24 | 2021-07-01 | 华为技术有限公司 | Neural network training method and device |
| CN111781244A (en) * | 2020-07-15 | 2020-10-16 | 中国科学院自动化研究所 | Infrared thermal imaging type coating detection method based on long-term and short-term memory network |
| CN111781244B (en) * | 2020-07-15 | 2021-10-26 | 中国科学院自动化研究所 | Infrared thermal imaging type coating detection method based on long-term and short-term memory network |
| CN113536655A (en) * | 2021-04-07 | 2021-10-22 | 北京聚树核科技有限公司 | Artificial intelligent deviation rectifying method and device for heliostat, electronic equipment and storage medium |
| CN114992138A (en) * | 2022-06-13 | 2022-09-02 | 湘潭大学 | Method for diagnosing faults of water supply pump of industrial circulating water system based on GAN-CNN-BilSTM |
| CN114992138B (en) * | 2022-06-13 | 2024-03-26 | 湘潭大学 | GAN-CNN-BiLSTM-based industrial circulating water system water supply pump fault diagnosis method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108562811B (en) | 2020-05-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108562811A (en) | Complicated electrical energy power quality disturbance analysis method based on the memory of two-way shot and long term | |
| CN112327219B (en) | Distribution transformer fault diagnosis method with automatic feature mining and parameter automatic optimizing functions | |
| CN104819846B (en) | Rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and sparse laminated automatic encoder | |
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| CN109828304B (en) | Deep learning-based method for predicting lithologic sequence model by using seismic data | |
| CN103886405B (en) | Boiler combustion condition identification method based on information entropy characteristics and probability nerve network | |
| CN108921285A (en) | Single-element classification method in sequence based on bidirectional valve controlled Recognition with Recurrent Neural Network | |
| CN113707176B (en) | Transformer fault detection method based on acoustic signal and deep learning technology | |
| CN107256393A (en) | The feature extraction and state recognition of one-dimensional physiological signal based on deep learning | |
| CN112581979B (en) | Speech emotion recognition method based on spectrogram | |
| CN103439647A (en) | Analog circuit fault diagnosis method for dynamic supply current monitoring | |
| CN107909118A (en) | A kind of power distribution network operating mode recording sorting technique based on deep neural network | |
| CN113609955A (en) | Three-phase inverter parameter identification method and system based on deep learning and digital twinning | |
| CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
| CN106773693A (en) | A kind of sparse causality analysis method of Industry Control multi-loop oscillation behavior | |
| CN110866448A (en) | Flutter signal analysis method based on convolutional neural network and short-time Fourier transform | |
| CN108535635A (en) | A kind of analog circuit intermittent fault diagnostic method based on EEMD and HMM | |
| CN110516742A (en) | A kind of distribution terminal fault distinguishing method and system based on Combination neural network model | |
| CN111368892A (en) | Generalized S transformation and SVM electric energy quality disturbance efficient identification method | |
| CN113884844A (en) | Transformer partial discharge type identification method and system | |
| CN117454100A (en) | Missing data reconstruction method for engineering structure | |
| Pawlak et al. | Nonparametric sequential signal change detection under dependent noise | |
| CN103577877B (en) | A kind of ship motion forecasting procedure based on time frequency analysis and BP neutral net | |
| CN116399592A (en) | Bearing fault diagnosis method based on channel attention dual-path feature extraction | |
| CN106096634A (en) | Fault detection method based on Adaptive windowing mental arithmetic method with interval halving algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |