CN110210152A - A kind of ultra harmonics source modeling method - Google Patents
A kind of ultra harmonics source modeling method Download PDFInfo
- Publication number
- CN110210152A CN110210152A CN201910494902.8A CN201910494902A CN110210152A CN 110210152 A CN110210152 A CN 110210152A CN 201910494902 A CN201910494902 A CN 201910494902A CN 110210152 A CN110210152 A CN 110210152A
- Authority
- CN
- China
- Prior art keywords
- output
- neuron
- ultra harmonics
- layer
- ultra
- 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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/36—Circuit design at the analogue level
- G06F30/367—Design verification, e.g. using simulation, simulation program with integrated circuit emphasis [SPICE], direct methods or relaxation methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Microelectronics & Electronic Packaging (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Geometry (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The present invention relates to a kind of ultra harmonics source modeling methods, including step S1: providing the ultra harmonics current time sequence data under the different capacity in any time period;Step S2: it is trained using data of the neural network algorithm to step S1;Step S3: the prediction data of each secondary ultra harmonics current amplitude under different capacity is generated using neural network algorithm;Step S4: carrying out error calculation for the step S3 prediction data generated and measured value, according to error calculation as a result, choosing the smallest performance evaluation coefficients R of error2, select the R under given computational accuracy ε2It is worth matched curve as fitting result;Step S5: analyzing and determining the final fitting result of step S4, if the error of obtained predicted value and measured value within the scope of, obtained trained neural network model.The present invention can predict the output ultra harmonics current characteristics in any one ultra harmonics source, to take necessary control measures or more suitable filter to be filtered in advance.
Description
Technical field
The present invention relates to field of power electronics, especially a kind of ultra harmonics source modeling method.
Background technique
With the fast development and energy-saving and environmental protection of the industries such as modernization industry, traffic, finance and information
The using energy source of the implementation of policy, clean and effective becomes mainstream.The high density access of a large amount of distributed energies and flexible transmission
Technology is widely applied so that power distribution network shows " source-net-lotus " close-coupled characteristic.When electric system to non-linear equipment and
When load is powered, fundamental wave energy that these non-linear equipments and load are supplied in transmitting, transformation, absorption system generator it is same
When, and part fundamental wave energy is converted to harmonic energy, foldback telegram in reply Force system becomes the main ultra harmonics source of power grid.
Currently, the infiltration with electrical equipments such as wind-powered electricity generation converter, photovoltaic DC-to-AC converter, electric automobile charging piles in power grid
Rate is higher and higher, generate ultra harmonics cause more and more power quality problems such as cause equipment intermittent work or
Disabler, equipment can not be worked or be damaged, power line carrier, PLC failure, device issue noise etc., these harm exist
It in the future undoubtedly can be increasingly severe.So the ultra harmonics propagation characteristic in power grid is accurately estimated, it is definite to grasp
The actual state of ultra harmonics in power grid safeguards that the safe operation of power grid is very necessary for preventing ultra harmonics from endangering.
Most common Harmonic Source Modeling method is mainly include the following types: (1) constant current source model and Nuo Dun Equivalent Model:
Although the class model is easily handled, but too simple rough, easily causes large error.And atypical operation is not accounted for
Situation reduces the precision and use scope of model.(2) simplified model based on crossover frequency admittance matrix: although the model
It is more to have considered mechanism that harmonic wave generates but proved that harmonic source characteristic linearize at a certain operating point without theoretical foundation,
And this model corresponds only to a certain operating point, there is different Harmonic source models for different operating points.(3) for common electricity
The topological circuit structure of power electronic device establishes harmonic current expression formula model, obtains it in exchange side by Fourier analysis
The individual harmonic current amplitude and phase angle of generation, model accuracy is high but calculates complexity, is only applicable in a certain specific topology,
There is no general applicability.Above mentioned model foundation this assumes that harmonic wave source structure is constant, when structure occurs one
When slight change, changes will occur for relationship, must just re-start analysis modeling.
Summary of the invention
In view of this, can predict any one the purpose of the present invention is to propose to a kind of ultra harmonics source modeling method
The output ultra harmonics current characteristics in ultra harmonics source, to take necessary control measures or selection more adduction in advance
Suitable filter is filtered.
The present invention is realized using following scheme: a kind of ultra harmonics source modeling method, comprising the following steps:
Step S1: the ultra harmonics current time sequence data under the different capacity in any time period is provided;
Step S2: using neural network algorithm to the ultra harmonics current time sequence data under the different capacity into
Row training;
Step S3: the prediction number of each secondary ultra harmonics current amplitude under different capacity is generated using neural network algorithm
According to;
Step S4: will be super under the different capacity being previously mentioned in the step S3 prediction data generated and measured value i.e. step S1
Higher harmonic current time series data carries out error calculation, according to error calculation as a result, choosing the smallest performance evaluation of error
Coefficients R2, select the R under given computational accuracy ε2It is worth matched curve as fitting result, to obtain the output of ultra harmonics source
Characteristic;
Step S5: analyzing and determining the final fitting result of step S4, if the mistake of obtained predicted value and measured value
Difference has then obtained trained neural network model within 0.0001-0.000001, defeated to trained neural network model
Enter the performance number and frequency in ultra harmonics source, which exports the ultra harmonics source of the required frequency under the power
Export current amplitude predicted value.
Further, the step S2 specifically includes the following steps:
Step S21: the corresponding ultra harmonics electric current Value Data of different frequency under different capacity is provided, comprising: input layer
Input vector, hidden layer input vectorHidden layer output vectorOutput layer input vectorWith the output of output layer
Vector;
Wherein, the input vector of input layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd it is super
Higher hamonic wave source measured current amplitude
The output vector of output layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd superelevation time
Harmonic source predicted current amplitude
The foundation of neural network model is carried out, input layer there are 2 nodes, and 1 node of output layer utilizes empirical equation: s=
2n+1, wherein n is input layer number, and calculating hidden layer node number is 5;It is gradually to increase on 5 number of nodes in hidden layer
Supernumerary segment points, are further added by number of nodes network error and no longer reduce when increasing to 10, the number of node selection at this time is optimal.
Step S22: to the connection weight ω of input layer and hidden layerih, hidden layer and output layer connection weight ωho, it is hidden
The threshold value b of each neuron containing layerh, each neuron of output layer threshold value boThe random number in one [- 11] is enabled respectively, and enables error
Function are as follows:Given computational accuracy value ε range is that 0.0001-0.000001 is terminated as training
Condition, maximum study number M range are 1000-1500;Learning rate selection range is 0.01-0.8;
Step S23: humorous to the corresponding superelevation time of different frequency under the different capacity in step S21 using S type activation primitive
Wave electric current Value Data is normalized to [01];
Step S24: the corresponding superelevation of different frequency under the different capacity in one group of input sample i.e. step S21 is randomly selected
Subharmonic current Value Data x (k) and corresponding desired output d (k);
Wherein, the ultra harmonics electric current Value Data x (k) includes the power in ultra harmonics sourceUltra harmonics
FrequencyAnd ultra harmonics source measured current amplitude
Step S25: the input hi of each neuron of hidden layer is calculatedh(k), it then uses input and swashs
Function living calculates the output ho of each neuron of hidden layerh(k):
In formula, ωihThe weight of each neuron of hidden layer, x are directed toward for each neuron of input layeri(k) in kth group data
Data on i-th of neuron, bhFor the threshold value of each neuron of hidden layer;
The input hi of each neuron of hidden layerh(k) pass through Sigmoid type function:
Mapping obtain the output ho of each neuron of hidden layerh(k)
hoh(k)=f (hih(k))
Step S26: the input yi of each neuron of output layer is calculatedo(k), output then is calculated with input and activation primitive
The output yo of each neuron of layero(k):
In formula, ωhoThe weight of each neuron of output layer, ho are directed toward for each neuron of hidden layerh(k) each for hidden layer
The output of neuron, boFor the threshold value of neuron each in output layer;
The input yi of each neuron of output layero(k) pass through Sigmoid type function:Map
The output yo of each neuron of output layer outo(k)
yoo(k)=f (yio(k))
Step S27: error function is calculated to the partial derivative δ of each neuron of output layero(k):
δo(k)=(do(k)-yoo(k))yoo(k)(1-yoo(k))
Step S28: the connection weight ω of hidden layer to output layer is utilizedho(k), error function is to each neuron of output layer
Partial derivative δo(k) and the output ho of hidden layerh(k) error function is calculated to the partial derivative δ of each neuron of hidden layerh(k):
Step S29: using error function to the partial derivative δ of each neuron of output layero(k) defeated with each neuron of hidden layer
Ho outh(k) Lai Xiuzheng connection weight ωho(k) and threshold value bo(k):
For connection weight adjusted,For adjustment before connection weight,For threshold adjusted
Value,For the threshold value before adjustment, η is learning rate, the value between (0,1);
Step S210: using error function to the partial derivative δ of each neuron of hidden layerh(k) and input layer each neuron
Input xi(k) connection weight and threshold value are corrected;
For connection weight adjusted,For adjustment before connection weight,It is adjusted
Threshold value,For the threshold value before adjustment, η is learning rate, the value between (0,1);
Step S211: being the power for including ultra harmonics source according to output layer output vectorUltra harmonics frequencyAnd ultra harmonics source predicted current amplitudeWith corresponding desired output do(k) global error E is calculated:
Step S212: judging whether network error meets the requirements, when E < ε or study number are greater than the maximum times set
M, then algorithm terminates;Otherwise, next learning sample and corresponding desired output are randomly selected, step S24 is returned to, under
The learning process of one wheel.
Further, the step S4 specifically includes the following steps:
Step S41: if algorithm terminates, renormalization is carried out to true value to the output data of neural network algorithm result;
Step S42: evaluating the predicted value of ultra harmonics ource electric current amplitude, according to neural network performance evaluation system
Number R2:
For the ultra harmonics current amplitude predicted value of i-th of data;yiFor the ultra harmonics electric current of i-th of data
Amplitude true value;N is data amount check to select final fitting result curve and then obtain ultra harmonics source output characteristics;For
Neural network performance evaluation coefficients R2Choosing value, R2For value closer to 1, the effect of training is better, models fitting it is more preferable, can
Obtain trained neural network model;Therefore choose the R under given computational accuracy ε2The matched curve of value as fitting result, into
And obtain ultra harmonics source output characteristics.
Further, the step S5 specifically includes the following steps:
Step S51: to the power in trained neural network model input data ultra harmonics source in step S4With ultra harmonics frequency
Step S52: the trained neural network model output dataThat is ultra harmonics source refers in step s 51
Determine the predicted current amplitude of the specific frequency inputted in the step S51 under power;
Step S53: prediction ultra harmonics source output characteristics is obtained.
Compared with prior art, the invention has the following beneficial effects:
(1) present invention can predict the output ultra harmonics current characteristics in any one ultra harmonics source, so as to pre-
Necessary control measures or the more suitable filter of selection are first taken to be filtered.
(2) present invention realizes complicated to some topological structures, service condition variation harmonic source and analyzes, when training
Between less, precision it is high, can dynamic modeling, be the effective ways of Harmonic Source Modeling.
Detailed description of the invention
Fig. 1 is that the embodiment of the present invention is that electric automobile charging pile exports ultra harmonics current spectrum under different capacity
Figure.
Fig. 2 is that the embodiment of the present invention is that electric automobile charging pile exports ultra harmonics current spectrum figure under certain power.
Fig. 3 is that the embodiment of the present invention is ultra harmonics source performance prediction flow chart.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and embodiments.
As shown in figure 3, present embodiments providing a kind of ultra harmonics source modeling method, comprising the following steps:
Step S1: the superelevation under the different capacity that can represent ultra harmonics source operating condition in any time period is provided
Subharmonic current time series data;
Step S2: using neural network algorithm to the ultra harmonics current time sequence data under the different capacity into
Row training;
Step S3: the prediction number of each secondary ultra harmonics current amplitude under different capacity is generated using neural network algorithm
According to;
Step S4: will be super under the different capacity being previously mentioned in the step S3 prediction data generated and measured value i.e. step S1
Higher harmonic current time series data (or simulation value) carries out error calculation, according to error calculation as a result, choosing error minimum
Performance evaluation coefficients R2, select the R under given computational accuracy ε2It is worth matched curve as fitting result, to obtain superelevation time
Harmonic source output characteristics;Performance evaluation coefficients R2Value, R2Value shows the variable in the equation in step S42 closer to 1(i-th
The ultra harmonics current amplitude predicted value of a data) to yi(the ultra harmonics current amplitude true values of i-th of data)
Interpretability is stronger, and the effect of training is better, this model is fitted data more preferable;
Step S5: analyzing and determining the final fitting result of step S4, if the mistake of obtained predicted value and measured value
Difference has then obtained trained neural network model within 0.0001-0.000001, defeated to trained neural network model
Enter the performance number and frequency in ultra harmonics source, which exports the ultra harmonics source of the required frequency under the power
Export current amplitude predicted value.
In the present embodiment, the step S2 specifically includes the following steps:
Step S21: the corresponding ultra harmonics electric current Value Data of different frequency under different capacity is provided, comprising: input layer
Input vector, hidden layer input vectorHidden layer output vectorOutput layer input vectorWith the output of output layer
Vector;
Wherein, the input vector of input layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd it is super
Higher hamonic wave source measured current amplitude
The output vector of output layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd superelevation time
Harmonic source predicted current amplitude
The foundation of neural network model is carried out, input layer there are 2 nodes, and 1 node of output layer utilizes empirical equation: s=
2n+1, wherein n is input layer number, and calculating hidden layer node number is 5;It is gradually to increase on 5 number of nodes in hidden layer
Supernumerary segment points, are further added by number of nodes network error and no longer significantly reduce when increasing to 10, the number of node selection at this time is optimal.
Step S22: to the connection weight ω of input layer and hidden layerih, hidden layer and output layer connection weight ωho, it is hidden
The threshold value b of each neuron containing layerh, each neuron of output layer threshold value boThe random number in one [- 11] is enabled respectively, and enables error
Function are as follows:Given computational accuracy value ε range is that 0.0001-0.000001 is terminated as training
Condition, maximum study number M range are 1000-1500;Learning rate selection range is 0.01-0.8;
Step S23: humorous to the corresponding superelevation time of different frequency under the different capacity in step S21 using S type activation primitive
Wave electric current Value Data is normalized to [01];
Step S24: the corresponding superelevation of different frequency under the different capacity in one group of input sample i.e. step S21 is randomly selected
Subharmonic current Value Data x (k) and corresponding desired output d (k);
Wherein, the ultra harmonics electric current Value Data x (k) includes the power in ultra harmonics sourceUltra harmonics
FrequencyAnd ultra harmonics source measured current amplitude
Step S25: the input hi of each neuron of hidden layer is calculatedh(k), it is then calculated with input and activation primitive implicit
The output ho of each neuron of layerh(k):
In formula, ωihThe weight of each neuron of hidden layer, x are directed toward for each neuron of input layeri(k) in kth group data
Data on i-th of neuron, bhFor the threshold value of each neuron of hidden layer;
The input hi of each neuron of hidden layerh(k) pass through Sigmoid type function:
Mapping obtain the output ho of each neuron of hidden layerh(k)
hoh(k)=f (hih(k))
Step S26: the input yi of each neuron of output layer is calculatedo(k), output then is calculated with input and activation primitive
The output yo of each neuron of layero(k):
In formula, ωhoThe weight of each neuron of output layer, ho are directed toward for each neuron of hidden layerh(k) each for hidden layer
The output of neuron, boFor the threshold value of neuron each in output layer;
The input yi of each neuron of output layero(k) pass through Sigmoid type function:Map
The output yo of each neuron of output layer outo(k)
yoo(k)=f (yio(k))
Step S27: error function is calculated to the partial derivative δ of each neuron of output layero(k):
δo(k)=(do(k)-yoo(k))yoo(k)(1-yoo(k))
Step S28: the connection weight ω of hidden layer to output layer is utilizedho(k), error function is to each neuron of output layer
Partial derivative δo(k) and the output ho of hidden layerh(k) error function is calculated to the partial derivative δ of each neuron of hidden layerh(k):
Step S29: using error function to the partial derivative δ of each neuron of output layero(k) defeated with each neuron of hidden layer
Ho outh(k) Lai Xiuzheng connection weight ωho(k) and threshold value bo(k):
For connection weight adjusted,For adjustment before connection weight,It is adjusted
Threshold value,For the threshold value before adjustment, η is learning rate, the value between (0,1).
Step S210: using error function to the partial derivative δ of each neuron of hidden layerh(k) and input layer each neuron
Input xi(k) connection weight and threshold value are corrected;
For connection weight adjusted,For adjustment before connection weight,For threshold adjusted
Value,For the threshold value before adjustment, η is learning rate, the value between (0,1).
Step S211: being the power for including ultra harmonics source according to output layer output vectorUltra harmonics frequencyAnd ultra harmonics source predicted current amplitudeWith corresponding desired output do(k) global error E is calculated:
Step S212: judging whether network error meets the requirements, when E < ε or study number are greater than the maximum times set
M, then algorithm terminates;Otherwise, next learning sample and corresponding desired output are randomly selected, step S24 is returned to, under
The learning process of one wheel.
In the present embodiment, the step S4 specifically includes the following steps:
Step S41: if algorithm terminates, renormalization is carried out to true value to the output data of neural network algorithm result;
Step S42: evaluating the predicted value of ultra harmonics ource electric current amplitude, according to neural network performance evaluation system
Number R2:
For the ultra harmonics current amplitude predicted value of i-th of data;yiFor the ultra harmonics electric current of i-th of data
Amplitude true value;N is data amount check to select final fitting result curve and then obtain ultra harmonics source output characteristics;For
Neural network performance evaluation coefficients R2Choosing value, R2For value closer to 1, the effect of training is better, models fitting it is more preferable, can
Obtain trained neural network model;Therefore choose the R under given computational accuracy ε2The matched curve of value as fitting result, into
And obtain ultra harmonics source output characteristics.
In the present embodiment, the step S5 specifically includes the following steps:
Step S51: to the power in trained neural network model input data ultra harmonics source in step S4With ultra harmonics frequency
Step S52: the trained neural network model output dataThat is ultra harmonics source refers in step s 51
Determine the predicted current amplitude of the specific frequency inputted in the step S51 under power;
Step S53: prediction ultra harmonics source output characteristics is obtained.
Preferably, the specific implementation process of the present embodiment is as follows: key step is as follows:
(1) superelevation time obtained from ultra harmonics source output terminal in a period of time acquired by electric energy quality test instrument is humorous
Wave current data, or the current data that emulation obtains, and form time series data;
(2) the ultra harmonics current time sequence data under different capacity is trained using neural network algorithm;
(3) a series of prediction of each secondary ultra harmonics current amplitude under different capacities is generated using neural network algorithm
Data;
(4) prediction data of generation and measured value (or simulation value) are subjected to error calculation, according to error calculation as a result, choosing
Take the smallest performance evaluation coefficients R of error2(performance evaluation coefficients R2Value, R2Value is closer to 1, and the effect of training is better, model
What is be fitted is more preferable) size, select suitable R2The matched curve of value obtains the output of ultra harmonics source as fitting result
Characteristic;
(5) to the performance number and frequency in trained neural network model input ultra harmonics source, model output
The output current amplitude predicted value in the ultra harmonics source of the required frequency under the power.
It is illustrated in fig. 1 shown below, obtains the output ultra harmonics current value of electric automobile charging pile under different capacity.Fig. 2 is
Ultra harmonics current value under a certain power situation, by it with the ultra harmonics source that is obtained by neural network algorithm in phase
It is compared with the harmonic prediction value under power, selects R2Matched curve when=0.95433 is carried out as fitting result
Analysis, and then obtain the output characteristics of electric automobile charging pile ultra harmonics electric current.
The present embodiment by under different capacity ultra harmonics source actual measurement data or the obtained data of emulation into
Method row training and then establish model.Ultra harmonics electric current output of the ultra harmonics source under certain power situation is obtained
Spectral characteristic.
The ultra harmonics current data (may be from largely surveying or emulating) that the present embodiment exports ultra harmonics source
It is trained, the output data of network is compared with real data, and change weight according to learning rules, until network is defeated
The difference of data and real data reaches within the error range of requirement out, to establish ultra harmonics model.
The foregoing is merely presently preferred embodiments of the present invention, all equivalent changes done according to scope of the present invention patent with
Modification, is all covered by the present invention.
Claims (4)
1. a kind of ultra harmonics source modeling method, it is characterised in that: the following steps are included:
Step S1: the ultra harmonics current time sequence data under the different capacity in any time period is provided;
Step S2: the ultra harmonics current time sequence data under the different capacity is instructed using neural network algorithm
Practice;
Step S3: the prediction data of each secondary ultra harmonics current amplitude under different capacity is generated using neural network algorithm;
Step S4: by the superelevation time under the different capacity being previously mentioned in the step S3 prediction data generated and measured value i.e. step S1
Harmonic current time series data carries out error calculation, according to error calculation as a result, choosing the smallest performance evaluation coefficient of error
R2, select the R under given computational accuracy ε2It is worth matched curve as fitting result, to obtain ultra harmonics source output characteristics;
Step S5: analyzing and determining the final fitting result of step S4, if the error of obtained predicted value and measured value exists
Within 0.0001-0.000001, then trained neural network model is obtained, trained neural network model has been inputted super
The performance number and frequency in higher hamonic wave source, the model export the output in the ultra harmonics source of the required frequency under the power
Current amplitude predicted value.
2. a kind of ultra harmonics source modeling method according to claim 1, it is characterised in that: the step S2 is specifically wrapped
Include following steps:
Step S21: provide different capacity under the corresponding ultra harmonics electric current Value Data of different frequency, comprising: input layer it is defeated
Incoming vector, hidden layer input vectorHidden layer output vectorOutput layer input vectorWith the output vector of output layer;
Wherein, the input vector of input layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd superelevation time
Harmonic source measured current amplitude
The output vector of output layer includes the power in ultra harmonics sourceUltra harmonics frequencyAnd ultra harmonics
Source predicted current amplitude
The foundation of neural network model is carried out, input layer there are 2 nodes, and 1 node of output layer utilizes empirical equation: s=2n+
1, wherein n is input layer number, and calculating hidden layer node number is 5;It is to gradually increase section on 5 number of nodes in hidden layer
Points, are further added by number of nodes network error and no longer reduce when increasing to 10, the number of node selection at this time is optimal.
Step S22: to the connection weight ω of input layer and hidden layerih, hidden layer and output layer connection weight ωho, hidden layer
The threshold value b of each neuronh, each neuron of output layer threshold value boThe random number in one [- 11] is enabled respectively, and enables error function
Are as follows:Given computational accuracy value ε range be 0.0001-0.000001 as training termination condition,
Maximum study number M range is 1000-1500;Learning rate selection range is 0.01-0.8;
Step S23: using S type activation primitive to the corresponding ultra harmonics electricity of different frequency under the different capacity in step S21
Flow valuve data are normalized to [0 1];
Step S24: it is humorous to randomly select the corresponding superelevation time of different frequency under the different capacity in one group of input sample i.e. step S21
Wave electric current Value Data x (k) and corresponding desired output d (k);
Wherein, the ultra harmonics electric current Value Data x (k) includes the power in ultra harmonics sourceUltra harmonics frequencyAnd ultra harmonics source measured current amplitude
Step S25: the input hi of each neuron of hidden layer is calculatedh(k), then each with input and activation primitive calculating hidden layer
The output ho of neuronh(k):
In formula, ωihThe weight of each neuron of hidden layer, x are directed toward for each neuron of input layeriIt (k) is i-th in kth group data
Data on a neuron, bhFor the threshold value of each neuron of hidden layer;
The input hi of each neuron of hidden layerh(k) pass through Sigmoid type function:
Mapping obtain the output ho of each neuron of hidden layerh(k)
hoh(k)=f (hih(k))
Step S26: the input yi of each neuron of output layer is calculatedo(k), then each with input and activation primitive calculating output layer
The output yo of neurono(k):
In formula, ωhoThe weight of each neuron of output layer, ho are directed toward for each neuron of hidden layerhIt (k) is each nerve of hidden layer
The output of member, boFor the threshold value of neuron each in output layer;
The input yi of each neuron of output layero(k) pass through Sigmoid type function:Mapping obtain it is defeated
The output yo of each neuron of layer outo(k)
yoo(k)=f (yio(k))
Step S27: error function is calculated to the partial derivative δ of each neuron of output layero(k):
δo(k)=(do(k)-yoo(k))yoo(k)(1-yoo(k))
Step S28: the connection weight ω of hidden layer to output layer is utilizedho(k), local derviation of the error function to each neuron of output layer
Number δo(k) and the output ho of hidden layerh(k) error function is calculated to the partial derivative δ of each neuron of hidden layerh(k):
Step S29: using error function to the partial derivative δ of each neuron of output layero(k) and the output ho of each neuron of hidden layerh
(k) Lai Xiuzheng connection weight ωho(k) and threshold value bo(k):
For connection weight adjusted,For adjustment before connection weight,For threshold value adjusted,For the threshold value before adjustment, η is learning rate, the value between (0,1);
Step S210: using error function to the partial derivative δ of each neuron of hidden layerh(k) and the input x of each neuron of input layeri
(k) connection weight and threshold value are corrected;
For connection weight adjusted,For adjustment before connection weight,For threshold value adjusted,For the threshold value before adjustment, η is learning rate, the value between (0,1);
Step S211: being the power for including ultra harmonics source according to output layer output vectorUltra harmonics frequency
And ultra harmonics source predicted current amplitudeWith corresponding desired output do(k) global error E is calculated:
Step S212: judging whether network error meets the requirements, and as E < ε or learns maximum times M of the number greater than setting, then
Algorithm terminates;Otherwise, next learning sample and corresponding desired output are randomly selected, step S24 is returned to, into next round
Learning process.
3. a kind of ultra harmonics source modeling method according to claim 1, it is characterised in that: the step S4 is specifically wrapped
Include following steps:
Step S41: if algorithm terminates, renormalization is carried out to true value to the output data of neural network algorithm result;
Step S42: evaluating the predicted value of ultra harmonics ource electric current amplitude, according to neural network performance evaluation coefficients R2:
For the ultra harmonics current amplitude predicted value of i-th of data;yiFor the ultra harmonics current amplitude of i-th of data
True value;N is data amount check to select final fitting result curve and then obtain ultra harmonics source output characteristics;For nerve
Network performance evaluation coefficients R2Choosing value, R2For value closer to 1, the effect of training is better, models fitting it is more preferable, can obtain
Trained neural network model;Therefore choose the R under given computational accuracy ε2The matched curve of value is obtained as fitting result
Obtain ultra harmonics source output characteristics.
4. a kind of ultra harmonics source modeling method according to claim 1, it is characterised in that: the step S5 is specifically wrapped
Include following steps:
Step S51: to the power in trained neural network model input data ultra harmonics source in step S4With it is super
Order harmonic frequencies
Step S52: the trained neural network model output dataI.e. function is specified in ultra harmonics source in step s 51
The predicted current amplitude of the specific frequency inputted in step S51 under rate;
Step S53: prediction ultra harmonics source output characteristics is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910494902.8A CN110210152B (en) | 2019-06-06 | 2019-06-06 | Ultrahigh-order harmonic source modeling method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910494902.8A CN110210152B (en) | 2019-06-06 | 2019-06-06 | Ultrahigh-order harmonic source modeling method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110210152A true CN110210152A (en) | 2019-09-06 |
CN110210152B CN110210152B (en) | 2021-01-29 |
Family
ID=67791519
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910494902.8A Active CN110210152B (en) | 2019-06-06 | 2019-06-06 | Ultrahigh-order harmonic source modeling method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110210152B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110021929A (en) * | 2018-12-29 | 2019-07-16 | 国网内蒙古东部电力有限公司经济技术研究院 | A kind of high-speed switch type fault current limiter electro-magnetic transient time-domain-simulation modeling method |
CN112531710A (en) * | 2020-11-18 | 2021-03-19 | 福州大学 | Method for predicting and evaluating harmonic source access |
CN113484605A (en) * | 2021-07-05 | 2021-10-08 | 华北电力大学 | Charger super-high harmonic emission evaluation method and device and storage medium |
Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101242101A (en) * | 2007-02-09 | 2008-08-13 | 上海磁浮交通工程技术研究中心 | Establishment method and simulation method for load harmony wave stabalized status model of power system |
CN101441231A (en) * | 2008-09-27 | 2009-05-27 | 武汉大学 | Harmonic electric energy metering error analytical apparatus |
US20110046924A1 (en) * | 2009-08-24 | 2011-02-24 | International Business Machines Corporation | Method for joint modeling of mean and dispersion |
CN103424621A (en) * | 2013-08-20 | 2013-12-04 | 江苏大学 | Artificial neural network detecting method of harmonic current |
CN103745065A (en) * | 2014-01-17 | 2014-04-23 | 国家电网公司 | Method for determining current after electrified railway multi-harmonic source superposition |
US20140351188A1 (en) * | 2013-05-22 | 2014-11-27 | Exablade Corporation | Prefetch system and method |
US20140362617A1 (en) * | 2013-06-07 | 2014-12-11 | The Board Of Trustees Of The University Of Alabama | Vector control of grid-connected power electronic converter using artificial neural networks |
CN104951836A (en) * | 2014-03-25 | 2015-09-30 | 上海市玻森数据科技有限公司 | Posting predication system based on nerual network technique |
US20150331062A1 (en) * | 2014-05-16 | 2015-11-19 | Boe Technology Group Co., Ltd. | Failure Detection Method and Detection Device for Inverter |
CN105550450A (en) * | 2015-12-17 | 2016-05-04 | 国网智能电网研究院 | Modeling method of electric energy quality interference source characteristic harmonic wave |
CN106849705A (en) * | 2017-03-24 | 2017-06-13 | 三峡大学 | A kind of current predictive control method based on single-phase Vienna rectifiers |
CN107147142A (en) * | 2017-05-12 | 2017-09-08 | 天津大学 | It is a kind of count and probabilistic distributed power source as harmonic source modeling method |
CN107769259A (en) * | 2017-12-01 | 2018-03-06 | 北京航空航天大学 | A kind of inverter current forecast Control Algorithm based on discrete averaging model |
CN109800520A (en) * | 2019-01-25 | 2019-05-24 | 国网浙江省电力有限公司湖州供电公司 | A kind of electric automobile charging station Harmonic Modeling method neural network based |
-
2019
- 2019-06-06 CN CN201910494902.8A patent/CN110210152B/en active Active
Patent Citations (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101242101A (en) * | 2007-02-09 | 2008-08-13 | 上海磁浮交通工程技术研究中心 | Establishment method and simulation method for load harmony wave stabalized status model of power system |
CN101441231A (en) * | 2008-09-27 | 2009-05-27 | 武汉大学 | Harmonic electric energy metering error analytical apparatus |
US20110046924A1 (en) * | 2009-08-24 | 2011-02-24 | International Business Machines Corporation | Method for joint modeling of mean and dispersion |
US20140351188A1 (en) * | 2013-05-22 | 2014-11-27 | Exablade Corporation | Prefetch system and method |
US20140362617A1 (en) * | 2013-06-07 | 2014-12-11 | The Board Of Trustees Of The University Of Alabama | Vector control of grid-connected power electronic converter using artificial neural networks |
CN103424621A (en) * | 2013-08-20 | 2013-12-04 | 江苏大学 | Artificial neural network detecting method of harmonic current |
CN103745065A (en) * | 2014-01-17 | 2014-04-23 | 国家电网公司 | Method for determining current after electrified railway multi-harmonic source superposition |
CN104951836A (en) * | 2014-03-25 | 2015-09-30 | 上海市玻森数据科技有限公司 | Posting predication system based on nerual network technique |
US20150331062A1 (en) * | 2014-05-16 | 2015-11-19 | Boe Technology Group Co., Ltd. | Failure Detection Method and Detection Device for Inverter |
CN105550450A (en) * | 2015-12-17 | 2016-05-04 | 国网智能电网研究院 | Modeling method of electric energy quality interference source characteristic harmonic wave |
CN106849705A (en) * | 2017-03-24 | 2017-06-13 | 三峡大学 | A kind of current predictive control method based on single-phase Vienna rectifiers |
CN107147142A (en) * | 2017-05-12 | 2017-09-08 | 天津大学 | It is a kind of count and probabilistic distributed power source as harmonic source modeling method |
CN107769259A (en) * | 2017-12-01 | 2018-03-06 | 北京航空航天大学 | A kind of inverter current forecast Control Algorithm based on discrete averaging model |
CN109800520A (en) * | 2019-01-25 | 2019-05-24 | 国网浙江省电力有限公司湖州供电公司 | A kind of electric automobile charging station Harmonic Modeling method neural network based |
Non-Patent Citations (2)
Title |
---|
WENBO SU .ET AL: ""The harmontic calculation model of non-linear load"", 《CICED 2014》 * |
占勇等: ""基于广义生长-剪枝径向基函数神经网络的谐波源建模"", 《中国电机工程学报》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110021929A (en) * | 2018-12-29 | 2019-07-16 | 国网内蒙古东部电力有限公司经济技术研究院 | A kind of high-speed switch type fault current limiter electro-magnetic transient time-domain-simulation modeling method |
CN110021929B (en) * | 2018-12-29 | 2023-01-13 | 国网内蒙古东部电力有限公司经济技术研究院 | Electromagnetic transient time domain simulation modeling method for fast switch type fault current limiter |
CN112531710A (en) * | 2020-11-18 | 2021-03-19 | 福州大学 | Method for predicting and evaluating harmonic source access |
CN113484605A (en) * | 2021-07-05 | 2021-10-08 | 华北电力大学 | Charger super-high harmonic emission evaluation method and device and storage medium |
CN113484605B (en) * | 2021-07-05 | 2022-04-22 | 华北电力大学 | Charger super-high harmonic emission evaluation method and device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110210152B (en) | 2021-01-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | Methods to aggregate turbine and network impedance for wind farm resonance analysis | |
CN110210152A (en) | A kind of ultra harmonics source modeling method | |
Ahmadipour et al. | Islanding detection technique using Slantlet Transform and Ridgelet Probabilistic Neural Network in grid-connected photovoltaic system | |
CN105184027A (en) | Power load modeling method based on interactive multi-model algorithm | |
CN111308260B (en) | Electric energy quality monitoring and electric appliance fault analysis system based on wavelet neural network and working method thereof | |
CN110323745A (en) | A kind of analytic method of modularization multi-level converter alternating current-direct current side harmonic propagation characteristic | |
CN105656036B (en) | Consider trend and the probability static security analysis method of sensitivity uniformity equivalence | |
Sahu et al. | A protection method for multi-terminal HVDC system based on fuzzy approach | |
CN105354628A (en) | Robust available power transmission capacity evaluation method for power transmission system | |
CN102709911B (en) | Method for designing interface with harmonic characteristic hybrid simulation function | |
CN102087311A (en) | Method for improving measurement accuracy of power mutual inductor | |
Darabian et al. | A UPFC‐based robust damping controller for optimal use of renewable energy sources in modern renewable integrated power systems | |
Alkayyali et al. | Hybrid PSO–ANN algorithm to control TCR for voltage balancing | |
CN108847673A (en) | The Probabilistic Load Flow method based on NATAF transformation in the uncertain source of arbitrariness probability distributing is obeyed in a kind of consideration AC-DC hybrid power grid | |
CN109462245B (en) | Local area power grid negative sequence unbalance comprehensive treatment method based on least square method | |
CN106953331A (en) | A kind of Harmonic Power Flow of Power Systems algorithm for considering uncertain and three-phase imbalance | |
CN117335426A (en) | Key feature identification method suitable for short-term voltage stability evaluation | |
CN103401271A (en) | Method and device for determining accepting ability of provincial power grid to distributed power sources | |
Aali | Shunt active DC filter based on intelligence controller for HVDC link | |
CN113794198A (en) | Method, device, terminal and storage medium for suppressing broadband oscillation | |
CN113962162A (en) | Simulation modeling method for alternating current-direct current hybrid system containing high-proportion new energy | |
Yue et al. | A power system harmonic problem based on the BP neural network learning algorithm | |
Blanco et al. | Development of a laboratory test bench for the emulation of wave energy converters | |
CN103560529B (en) | Control method for restraining irregular power fluctuation of power system alternating current tie lines | |
Abd-EI et al. | Multi-machine power system dynamic equivalents using artificial intelligence (ANN) |
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 |