CN113189513B - Ripple-based redundant power supply current sharing state identification method - Google Patents
Ripple-based redundant power supply current sharing state identification method Download PDFInfo
- Publication number
- CN113189513B CN113189513B CN202110472729.9A CN202110472729A CN113189513B CN 113189513 B CN113189513 B CN 113189513B CN 202110472729 A CN202110472729 A CN 202110472729A CN 113189513 B CN113189513 B CN 113189513B
- Authority
- CN
- China
- Prior art keywords
- power supply
- power
- peak
- ripple
- power spectrum
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000001228 spectrum Methods 0.000 claims abstract description 71
- 238000003062 neural network model Methods 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims description 18
- 230000004913 activation Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 6
- 238000012887 quadratic function Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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
- G01R31/40—Testing power supplies
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R19/00—Arrangements for measuring currents or voltages or for indicating presence or sign thereof
- G01R19/0092—Arrangements for measuring currents or voltages or for indicating presence or sign thereof measuring current only
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R21/00—Arrangements for measuring electric power or power factor
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R23/00—Arrangements for measuring frequencies; Arrangements for analysing frequency spectra
- G01R23/02—Arrangements for measuring frequency, e.g. pulse repetition rate; Arrangements for measuring period of current or voltage
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a redundant power supply current-sharing state identification method based on ripples, which comprises the steps of firstly collecting the current and ripples of each power supply at different moments and the total ripples of the output end of a redundant module, and then calculating the power and unilateral power spectrum of each power supply at different moments and the unilateral power spectrum of the total ripples; then, acquiring the frequency at the maximum value of the unilateral power spectrum of each power supply ripple and the frequency corresponding to the peak value condition on the unilateral power spectrum of the total ripple; and then fitting a power curve, building and training a neural network model, predicting the probability value tensor of the power supply label through the neural network, and further completing the identification of the current sharing state of the redundant power supply.
Description
Technical Field
The invention belongs to the technical field of power supply state identification, and particularly relates to a redundant power supply current sharing state identification method based on ripples.
Background
The redundant power supply is used for supplying power to the server at first, and because the server needs to work stably for a long time, and the data loss of the server can be caused by sudden power failure, a plurality of power supplies are adopted to form the redundant power supply to supply power to the server, when one power supply is abnormal or stops working, other power supplies can supply power to the server, and the server can be ensured to be in a power supply state all the time. At present, no matter the computer mainboard supplies power, or the large-scale automatic control system supplies power, the redundant power supply structure is adopted to supply power, the stability of power supply in the operation process of the equipment is ensured, and unnecessary loss caused by power failure is prevented.
Generally, the switch-type redundant power supply can automatically adjust under the condition of load change, such as corresponding output current and the like, and monitoring and analysis of the switch-type redundant power supply are essential, sometimes, the working state of each power supply, including parameters such as output power and the like of each power supply, is necessary to know, which is important for the whole redundant system, and sometimes, whether the redundant power supply fails or not can be predicted from the parameters, and which power supply fails at the same time, so that the probability of advanced maintenance is greatly improved, and the system is prevented from being shut down.
According to the principle of the switching power supply, the main frequency components of the power spectrum of the output ripple are the switching frequency and the harmonic wave of the switching power supply, and meanwhile, the output power of the power supply has a certain relation with the frequency in a specific frequency band of the power spectrum. For different power supplies, due to the difference of the electrical characteristics, the waveforms of the peak parts of the power spectrum have certain difference. Since the whole redundant power supply should be load balanced during normal operation, but the operating load of the redundant power supply is unbalanced due to long-term operation, aging and the like, the frequency component in the power spectrum of the total ripple can reflect the unbalanced relation.
In the conventional method, parameters such as current and voltage are acquired for a single power supply in a redundant structure and are subjected to signal processing, so that the output condition of each power supply is determined, and the state monitoring is performed on the redundant structure, but the hardware cost and the limitation are increased unintentionally, for example, in a redundant power supply with good packaging, the state data of the single power supply cannot be acquired, and the monitoring on the single power supply is difficult to realize. The method focuses on knowing the output distribution condition of each power supply from the total ripple waves, and further analyzes whether the working load of the redundant power supply is balanced or not, so that the working condition of the power supply is reflected, the difficulty in researching a single power supply can be overcome, and meanwhile, the condition of each single power supply can be analyzed under the condition that only the total output is obtained.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a redundant power supply current-sharing state identification method based on ripples.
In order to achieve the above object, the present invention provides a method for identifying a current sharing state of a redundant power supply based on ripples, which is characterized by comprising the following steps:
(1) Inputting all power supplies in parallel to a redundancy module, and then collecting the current I of the output end of each power supply at m sampling moments i (t) and ripple W i (t) and total ripple W (t) at the output of the redundant module, where i =1,2, \8230;, k, k denotes power supply maximum number, t =1,2, \8230;, m;
(2) Calculating the output power P of each power supply at each sampling moment i (t)；
P i (t)＝(I i (t)) 2 R(t) (1)
Wherein, R (t) represents the load resistance value at the tth sampling moment;
(3) Acquiring the frequency of the maximum value of the unilateral power spectrum of each power supply ripple;
(3.1) acquiring a single-side power spectrum of each power supply ripple at each sampling moment by using fast Fourier transform;
where FFT is a fast Fourier transform function, conj is a conjugate function for calculating complex numbers, FFT i Is lineResult of fast Fourier transform of wave, dataP i (t) represents the single-sided power spectrum of the ith power supply ripple at the tth sampling instant;
(3.2) obtaining the frequency f at the maximum value of the single-side power spectrum of each power supply ripple i max (t)；
Wherein INDEXOF represents the position of the acquired corresponding value in the whole data;
(4) Searching a frequency corresponding to the condition meeting the peak value on the unilateral power spectrum of the total ripple;
(4.1) setting the minimum distance between two adjacent peak values on the single-side power spectrum of the total ripple wave as D; setting the threshold value of the peak value to A min ；
(4.2) acquiring a single-side power spectrum dataP (t) of the total ripple at each sampling moment according to the method of the formula (2);
(4.3) first finding the first conditional peak A on the unilateral power spectrum dataP (t) 1 (t)，A 1 (t) satisfies A 1 (t)＞A min Then record A 1 (t) corresponding frequency f 1 (t), and marking the current position as count;
(4.4) taking the current position count as a starting point, continuously traversing the unilateral power spectrum dataP (t) beyond the minimum distance D after the position count to find a second peak A meeting the condition 2 (t)，A 2 (t) also satisfies A 2 (t)＞A min Then record A 2 (t) corresponding frequency f 2 (t) and mixing f 2 (t) marking the position as the current position count;
(4.5) repeating the step (4.4), and finding all peak values A meeting the conditions on the unilateral power spectrum dataP (t) j (t) and recording the corresponding frequency f j (t)，j≤k；
(5) The output power P of each power supply i (t) and frequency f at which the single-sided power spectrum is at a maximum i max (t) in the following orderForming a data set;
data i ＝{(P i (1),f i max (1)),(P i (2),f i max (2)),...,(P i (m),f i max (m))} (4)
(6) For the ith data set data i Fitting a quadratic polynomial to obtain a curve L (f, p);
(6.1) setting a quadratic function model;
P(f)＝af 2 +bf+c (5)
wherein f represents frequency, P is corresponding power, and a, b and c are coefficients;
(6.2) data i Each group of frequency and corresponding power are brought into the formula (5) for fitting, and the fitting process meets the minimum mean square error w (a, b, c);
the extremum theorem yields:
the final goal is to solve the following system of equations:
(6.3) solving the equation set to calculate coefficients a, b and c, and substituting the coefficients a, b and c into the formula (5) to obtain a curve L (f, p);
(7) Acquiring peak waveform data;
(7.1) Single-sided Power Spectrum DataP at each Power supply ripple i (t) up to f i max (t) obtaining peak waveform data dataPeek with length n as center i (t)；
dataPeek i (t)＝dataP i (t)[f i max (t)-n/2,f i max (t)+n/2] (9)
(7.2) acquiring peak waveform data with the length n on the unilateral power spectrum dataP (t) of the total ripple;
(7.2.1) obtaining a first peak value A 1 (t) Peak waveform data DataPeek 1 (t)；
On the single-sided power spectrum dataP (t) of the total ripple, with f 1 (t) as a center, obtaining peak waveform data dataPeek with length n 1 (t)；
If f is 1 (t)+n/2＜f 2 (t), then dataPeek 1 (t)＝dataP(t)[f 1 (t)-n/2,f 1 (t)+n/2]；
If not, then,
then to
The following operations are carried out:
wherein INSERT (COUNT, 0, L) indicates that L0 are inserted in sequence at the COUNT position of corresponding data;
(7.2.2) obtaining the last peak value A last (t) Peak waveform data DataPeek last (t)；
Let the last peak A last (t) corresponds to a frequency f last (t); on the single-sided power spectrum dataP (t) of the total ripple, with f last (t) as a center, obtaining peak waveform data dataPeek with length n last (t)；
If f is last (t)-n/2＞f last-1 (t), then dataPeek last (t)＝dataP(t)[f last (t)-n/2,f last (t)+n/2]；
If not, then,
then to
The following operations are carried out:
(7.2.3) obtaining the jth peak value A j (t) Peak waveform data DataPeek j (t)，j＝2,3,…,(last-1)；
If f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
dataPeek j (t)＝dataP(t)[f j (t)-n/2,f j (t)+n/2]；
if f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＜f j-1 (t), then:
then to
The following operations are carried out:
if f is j (t)+n/2＞f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
then to
The following operations are carried out:
if none of the above conditions is met:
first to each other
The following operations are carried out:
then to
The following operations are carried out:
(8) Loading labels on all peak waveform data to construct an input data set V of the neural network;
wherein y is a tensor with the size of k × 1, the value of the position corresponding to the power supply number in the tensor of k × 1 is 1, and the values of other positions are 0;
(9) Building a neural network model;
the first layer is an input layer of the neural network, and the data dimension of the input layer is n x 1;
the second layer is a one-dimensional convolution layer, the number of convolution kernels is 128, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and average pooling are used for processing data simultaneously, the size of the pooled kernels is 3 x 1, and the step length is 1;
the third layer is a one-dimensional convolution layer, the number of convolution kernels is 64, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and maximum pooling are simultaneously used for processing, the size of the pooled kernel is 3 x 1, and the step length is 1;
the fourth layer is a one-dimensional convolution layer, the number of convolution kernels is 32, the size is 3 x 1, the step length is 1, and the activation function is a relu function;
the fifth layer is a flattening layer, and the convolution result is expanded into one-dimensional data;
the sixth layer is a fully-connected output layer, the activation function is softmax, and the output dimension is a probability value tensor of k × 1;
(10) Training a neural network model;
(10.1) selecting M groups of peak waveform data in batches from the input data set V and inputting the M groups of peak waveform data into the neural network model, thereby predicting probability value tensors of M groups of power supply labels
u＝1,2,…,M；
Wherein,
the l output of k outputs, a, representing the u-th set of peak waveform data through the neural network s The u-th set of peak waveform data is passed through the s-th output of the k outputs of the neural network,
a probability value representing that the u-th group of peak waveform data belongs to the l-th power source;
(10.2) then calculating a loss function value C of the neural network by the following formula;
wherein,
or
Indicating that the u-th group of peak waveform data belongs to the l-th power source in the category l,
indicating that the category l of the u group of peak waveform data does not belong to the l power supply;
(10.3) repeating the steps (10.1) and (10.2), and continuing to train the neural network model until the neural network model is converged to obtain the trained neural network model;
(11) Identifying the current sharing state of the redundant power supply;
(11.1) acquiring a total ripple single-sided power spectrum under the current load R (t), and then searching for the frequency f corresponding to the condition meeting the peak value on the total ripple single-sided power spectrum according to the method in the step (4) j (t)；
(11.2) acquiring peak waveform data dataPeek with the length of n on the unilateral power spectrum of the total ripple wave according to the method in the step (7.2) j (t)；
(11.3) mixing j (t) substituting into the fitted power curve L (f, P) to obtain the corresponding power P j (t)；
(11.4) preparing dapeek j (t) inputting the data to the trained neural network model, thereby predicting the probability value tensor of the power label
(11.5) according to
Calculating a corresponding power supply number;
(11.6) Number according to Power supply j And corresponding output power P j And (t) obtaining the ratio of the current of each power supply, thereby identifying whether each power supply reaches a current equalizing state.
The invention aims to realize the following steps:
the invention relates to a redundant power supply current-sharing state identification method based on ripples, which comprises the steps of firstly collecting the current and ripples of each power supply at different moments and the total ripples of the output end of a redundant module, and then calculating the power and unilateral power spectrum of each power supply at different moments and the unilateral power spectrum of the total ripples; then, acquiring the frequency at the maximum value of the unilateral power spectrum of each power supply ripple and the frequency corresponding to the peak value condition on the unilateral power spectrum of the total ripple; and then fitting a power curve, building and training a neural network model, predicting the probability value tensor of the power supply label through the neural network, and further completing the identification of the current sharing state of the redundant power supply.
Meanwhile, the ripple-based redundant power supply current sharing state identification method also has the following beneficial effects:
(1) The invention can not only extract the frequency of each power supply from the total ripple, but also calculate the corresponding power ratio by using the relationship between the frequency and the power so as to obtain the current ratio, thereby realizing a mapping relationship from the total ripple to the output of each power supply;
(2) The invention integrates a convolutional neural network model, maps the peak part waveform of the power supply with the power supply, and can know the power and the output power of each power supply, thereby providing a theoretical basis for subsequently providing redundant power supply control;
(3) The invention only uses the total output end ripple wave in the subsequent actual identification, and the data of each single power supply does not need to be measured when in use, thus playing the effects of simplifying hardware and the like on the monitoring of the redundant power supply and achieving single-point measurement;
(4) The method can perform real-time online analysis on the current sharing state of the redundant structure in a non-stop state, and only needs to fit a corresponding curve at the early stage and train to complete a corresponding model.
Drawings
FIG. 1 is a flow chart of a ripple-based method for identifying current sharing status of redundant power supplies according to the present invention;
FIG. 2 is a block diagram of one embodiment of a redundant power supply;
FIG. 3 is a diagram of ripple and a corresponding single-sided power spectrum for the first power supply;
FIG. 4 is a graph of total ripple and the corresponding single-sided power spectrum;
FIG. 5 is a power curve fitted to a first power supply;
FIG. 6 is a power curve fitted to a second power supply;
fig. 7 is a graph of predicted results of power and neural networks calculated by fitting curves to total ripple data.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
Examples
Fig. 1 is a flow chart of a method for identifying a current sharing state of a redundant power supply based on ripples according to the present invention.
In this embodiment, as shown in fig. 1, the method for identifying the current sharing status of the redundant power supply based on ripples according to the present invention includes the following steps:
s1, signal acquisition
As shown in FIG. 2, 2 power supplies are input to the redundancy module in parallel, and then the current I of each power supply output terminal is collected at 16 sampling moments i (t) and ripples W i (t), and a total ripple W (t) at the output of the redundant module, where i =1,2,3, k denotes the power supply maximum number, t =1,2, \8230;, 16;
s2, calculating the output power P of each power supply at each sampling moment i (t)；
P i (t)＝(I i (t)) 2 R(t) (1)
Wherein, R (t) represents the load resistance value at the tth sampling moment;
in this embodiment, the actual acquired power of the two power sources at different frequencies is shown in table 1;
TABLE 1
S3, acquiring the frequency of the maximum value of the unilateral power spectrum of each power supply ripple;
s3.1, acquiring a single-side power spectrum of each power supply ripple at each sampling moment by using fast Fourier transform;
where FFT is a fast Fourier transform function, conj is a conjugate function for calculating complex numbers, FFT i Is the result of fast Fourier transform of ripple, dataP i (t) represents the single-sided power spectrum of the ith supply ripple at the tth sampling instant;
s3.2, calculating the frequency f at the maximum value of the unilateral power spectrum of each power supply ripple i max (t)；
Wherein INDEXOF represents the position of the acquired corresponding value in the whole data;
in this embodiment, taking the first power supply as an example, the ripple of the first power supply is shown in (a) of fig. 3, and the corresponding half-side power spectrum is shown in a waveform diagram of 150kHz to 300kHz, as shown in (b) of fig. 3, when the power spectrum is negativeLoad of 10 Ω, corresponding to f 1 max (t) around 179 kHz.
S4, searching a frequency corresponding to a condition meeting a peak value on a single-side power spectrum of the total ripple;
s4.1, setting the minimum distance between two adjacent peak values on the single-side power spectrum of the total ripple wave as D; setting the threshold value of the peak value to A min ；
S4.2, acquiring a single-side power spectrum dataP (t) of the total ripple at each sampling moment according to the method of the formula (2);
s4.3, firstly finding the first peak A meeting the condition on the unilateral power spectrum dataP (t) 1 (t)，A 1 (t) satisfies A 1 (t)＞A min Then record A 1 (t) corresponding frequency f 1 (t), and marking the current position as count;
s4.4, with the current position count as a starting point, continuously traversing the unilateral power spectrum dataP (t) beyond the minimum distance D after the position count to find a second peak value A meeting the condition 2 (t)，A 2 (t) also satisfies A 2 (t)＞A min Then record A 2 (t) corresponding frequency f 2 (t) and mixing f 2 (t) marking the position as the current position count;
s4.5, repeating the step S4.4, and finding all peak values A meeting the conditions on the unilateral power spectrum dataP (t) j (t) and recording the corresponding frequency f j (t)，j≤k；
In this embodiment, the total ripple of the output of the redundant module is shown in (a) of fig. 4, and the corresponding half-side power spectrum is shown in a waveform diagram of 150kHz to 300kHz, as shown in (b) of fig. 4, and f is obtained at this time 1 (t),f 2 (t) around 200kHz and 235kHz, respectively.
S5, output power P of each power supply i (t) and frequency f at which the single-sided power spectrum is at a maximum i max (t) constructing a data set in the following order;
data i ＝{(P i (1),f i max (1)),(P i (2),f i max (2)),...,(P i (m),f i max (m))} (4)
s6, for the ith data set data i Fitting a quadratic polynomial to obtain a curve L (f, p);
s6.1, setting a quadratic function model;
P(f)＝af 2 +bf+c (5)
wherein f represents frequency, P is corresponding power, and a, b and c are coefficients;
s6.2, data i Each group of frequency and corresponding power are brought into the formula (5) for fitting, and the fitting process meets the minimum mean square error w (a, b, c);
the extremum theorem yields:
the final goal is to solve the following system of equations:
s6.3, solving the equation set to calculate coefficients a, b and c, and substituting the coefficients a, b and c into a curve which is fit by a formula (5) to obtain L (f, p);
in this embodiment, a curve fitted by the power supply 1 is shown in fig. 5, and the corresponding coefficients take values as follows: a =2.041170202001298e-09, b = -0.0015473741405332645, c = -268.9394698628 441136; the curve fitted by the power supply 2 is shown in fig. 6, and the corresponding coefficients take the values: a =2.1850582908645083e-09, b = -0.0016008335757875327, c = -273.2382862139963; the error of the fit of the two curves is 0.6 and 0.8, respectively.
S7, acquiring peak waveform data;
s7.1, single-side power spectrum dataP of each power supply ripple i (t) up to f i max (t) as a center, obtaining peak waveform data dataPeek with length n i (t)；
dataPeek i (t)＝dataP i (t)[f i max (t)-n/2,f i max (t)+n/2] (9)
S7.2, acquiring peak waveform data with the length n on the unilateral power spectrum dataP (t) of the total ripple;
s7.2.1, obtaining a first peak value A 1 (t) Peak waveform data DataPeek 1 (t)；
On the single-sided power spectrum dataP (t) of the total ripple, with f 1 (t) as a center, obtaining peak waveform data dataPeek with length n 1 (t)；
If f is 1 (t)+n/2＜f 2 (t), then dataPeek 1 (t)＝dataP(t)[f 1 (t)-n/2,f 1 (t)+n/2]；
If not, then the mobile terminal can be switched to the normal mode,
then to
The following operations are carried out:
wherein INSERT (COUNT, 0, L) indicates that L0 are inserted in sequence at the COUNT position of corresponding data;
s7.2.2, obtaining the last peak value A last (t) Peak waveform data DataPeek last (t)；
Let the last peak A last (t) corresponds to a frequency f last (t); on the single-sided power spectrum dataP (t) of the total ripple, with f last (t) obtaining peak waveform data dataPeek with length n as center last (t)；
If f is last (t)-n/2＞f last-1 (t), then dataPeek last (t)＝dataP(t)[f last (t)-n/2,f last (t)+n/2]；
If not, then the mobile terminal can be switched to the normal mode,
then to
The following operations are carried out:
s7.2.3, obtaining the jth peak value A j (t) Peak waveform data dataPeek j (t), j =2,3, \8230; (last-1); if f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
dataPeek j (t)＝dataP(t)[f j (t)-n/2,f j (t)+n/2]；
if f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＜f j-1 (t), then:
then to
The following operations are carried out:
if f is j (t)+n/2＞f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
then to
The following operations are carried out:
if none of the above conditions are met:
first to each other
The following operations are carried out:
then to
The following operations are carried out:
s8, loading labels on all peak waveform data, and constructing an input data set V of the neural network;
wherein y is a tensor with the size of k × 1, the value of the position corresponding to the power supply number in the tensor of k × 1 is 1, and the values of other positions are 0;
s9, building a neural network model;
the first layer is an input layer of the neural network, and the data dimension of the input layer is n x 1;
the second layer is a one-dimensional convolution layer, the number of convolution kernels is 128, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and average pooling are used for processing data simultaneously, the size of the pooled kernels is 3 x 1, and the step length is 1;
the third layer is a one-dimensional convolution layer, the number of convolution kernels is 64, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and maximum pooling are used for processing at the same time, the size of the pooled kernels is 3 x 1, and the step length is 1;
the fourth layer is a one-dimensional convolution layer, the number of convolution kernels is 32, the size is 3 x 1, the step length is 1, and the activation function is a relu function;
the fifth layer is a flattening layer, and the convolution result is expanded into one-dimensional data;
the sixth layer is a fully-connected output layer, the activation function is softmax, and the output dimension is a probability value tensor of k x 1;
s10, training a neural network model;
s10.1, selecting M groups of peak waveform data in batches from the input data set V and inputting the M groups of peak waveform data into a neural network model, thereby predicting probability value tensors of M groups of power labels
u＝1,2,…,M；
Wherein,
the l output of k outputs, a, representing the u-th set of peak waveform data through the neural network s The u-th set of peak waveform data is passed through the s-th output of the k outputs of the neural network,
a probability value representing that the u-th group of peak waveform data belongs to the l-th power source;
s10.2, calculating a loss function value C of the neural network through the following formula;
wherein,
or
Indicating that the u-th group of peak waveform data belongs to the l-th power source in the category l,
the method comprises the steps that the class l of the u-th group of peak waveform data does not belong to the ith power supply;
s10.3, repeating the steps S10.1 and S10.2, and continuing to train the neural network model until the neural network model converges to obtain the trained neural network model;
s11, identifying the current sharing state of the redundant power supply;
s11.1, acquiring a total ripple single-sided power spectrum under the current load R (t), and then searching a frequency f corresponding to the condition meeting the peak value on the single-sided power spectrum of the total ripple according to the method in the step S4 j (t)；
S11.2, acquiring peak waveform data dataPeek with length n on the unilateral power spectrum of the total ripple wave according to the method in the step S7.2 j (t)；
S11.3, mixing f j (t) substituting into the fitted power curve L (f, P) to obtain corresponding power P j (t)；
S11.4, mixing data Peek j (t) inputting the data to the trained neural network model, thereby predicting the probability value tensor of the power label
S11.5 according to
Calculating a corresponding power supply number;
s11.6, numbering according to the power supply j And corresponding output power P j And (t) obtaining the ratio of the current of each power supply, thereby identifying whether the redundant power supply reaches a current equalizing state.
Fig. 7 is a graph of predicted results of power and neural networks calculated by fitting curves to total ripple data.
In this embodiment, we show the method in a visual manner, and compare it with the actual output of each power source to verify the correctness of the method.
This example measured actual currents of 1.67A and 0.75A, respectively, and the present output voltage was 24.5, and the corresponding powers were 40.83W and 18.375W, respectively. Calculating the power of a first power supply to be 40.39W according to the fitted curve, measuring the power to be 40.83W, and obtaining the relative error to be 1.08%; calculating the power of a second power supply to be 18.67W according to the fitted curve, measuring the power to be 18.375W, and obtaining the relative error to be 1.58 percent; the ratio of the powers calculated from the fitted curve is 2.22:1, measured power ratio of 2.16:1. the above data illustrate the accuracy of this method, and the output power corresponding to each power supply can be obtained from the total ripple, and the ratio thereof can be obtained to indicate the current-sharing status.
Although the illustrative embodiments of the present invention have been described in order to facilitate those skilled in the art to understand the present invention, it is to be understood that the present invention is not limited to the scope of the embodiments, and that various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined in the appended claims, and all matters of the invention using the inventive concepts are protected.
Claims (1)
1. A redundant power supply current sharing state identification method based on ripples is characterized by comprising the following steps:
(1) Inputting all power supplies in parallel to the redundancy module, and thenCollecting current I of each power supply output end at m sampling moments i (t) and ripples W i (t), and a total ripple at the output of the redundant module W (t), where i =1,2, \8230, k denotes the power supply maximum number, t =1,2, \8230, m;
(2) Calculating the output power P of each power supply at each sampling moment i (t)；
P i (t)＝(I i (t)) 2 R(t) (1)
Wherein, R (t) represents the load resistance value at the t-th sampling moment;
(3) Acquiring the frequency of the maximum value of the unilateral power spectrum of each power supply ripple;
(3.1) acquiring a single-side power spectrum of each power supply ripple at each sampling moment by using fast Fourier transform;
where FFT is a fast Fourier transform function, conj is a conjugate function for calculating complex numbers, FFT i Is the fast Fourier transform result of the ripple, dataP i (t) represents the single-sided power spectrum of the ith power supply ripple at the tth sampling instant;
(3.2) obtaining the frequency f at the maximum value of the single-side power spectrum of each power supply ripple i max (t)；
Wherein INDEXOF represents the position of the acquired corresponding value in the whole data;
(4) Searching the frequency corresponding to the condition meeting the peak value on the unilateral power spectrum of the total ripple waves;
(4.1) setting the minimum distance between two adjacent peak values on the single-side power spectrum of the total ripple as D; setting the threshold value of the peak value to A min ；
(4.2) acquiring a single-side power spectrum dataP (t) of the total ripple at each sampling moment according to the method in the formula (2);
(4.3) first finding the first conditional peak A on the unilateral power spectrum dataP (t) 1 (t)，A 1 (t) satisfies A 1 (t)＞A min Then record A 1 (t) corresponding frequency f 1 (t), and marking the current position as count;
(4.4) taking the current position count as a starting point, continuously traversing the unilateral power spectrum dataP (t) beyond the minimum distance D after the position count to find a second peak A meeting the condition 2 (t)，A 2 (t) also satisfies A 2 (t)＞A min Then record A 2 (t) corresponding frequency f 2 (t) and mixing f 2 (t) marking the position of the current position as a count;
(4.5) repeating the step (4.4), and finding all peak values A meeting the condition on the unilateral power spectrum dataP (t) j (t) and recording the corresponding frequency f j (t)，j≤k；
(5) The output power P of each power supply i (t) and frequency f at which the single-sided power spectrum is at a maximum i max (t) constructing a data set in the following order;
data i ＝{(P i (1),f i max (1)),(P i (2),f i max (2)),...,(P i (m),f i max (m))} (4)
(6) For the ith data set data i Fitting a quadratic polynomial to obtain a curve L (f, p);
(6.1) setting a quadratic function model;
P(f)＝af 2 +bf+c (5)
wherein f represents frequency, P is corresponding power, and a, b and c are coefficients;
(6.2) data i Each group of frequencies and corresponding power are brought into a formula (5) for fitting, and the fitting process meets the minimum mean square error w (a, b, c);
the extremum theorem yields:
the final goal is to solve the following system of equations:
(6.3) solving the equation set to calculate coefficients a, b and c, and substituting the coefficients a, b and c into the formula (5) to obtain a curve L (f, p);
(7) Acquiring peak waveform data;
(7.1) Single-sided Power Spectrum DataP at each Power supply ripple i (t) up to f i max (t) as a center, obtaining peak waveform data dataPeek with length n i (t)；
dataPeek i (t)＝dataP i (t)[f i max (t)-n/2,f i max (t)+n/2] (9)
(7.2) acquiring peak waveform data with the length n on the unilateral power spectrum dataP (t) of the total ripple;
(7.2.1) obtaining a first peak value A 1 (t) Peak waveform data DataPeek 1 (t)；
On the single-sided power spectrum dataP (t) of the total ripple, with f 1 (t) as a center, obtaining peak waveform data dataPeek with length n 1 (t)；
If f is 1 (t)+n/2＜f 2 (t), then dataPeek 1 (t)＝dataP(t)[f 1 (t)-n/2,f 1 (t)+n/2]；
If not, then the mobile terminal can be switched to the normal mode,
then to
The following operations are carried out:
wherein INSERT (COUNT, 0, L) indicates that L0 are inserted in sequence at the COUNT position of corresponding data;
(7.2.2) obtaining the last peak value A last (t) Peak waveform data DataPeek last (t)；
Let the last peak A last (t) corresponds to a frequency f last (t); on the single-sided power spectrum dataP (t) of the total ripple, with f last (t) as a center, obtaining peak waveform data dataPeek with length n last (t)；
If f is last (t)-n/2＞f last-1 (t), then dataPeek last (t)＝dataP(t)[f last (t)-n/2,f last (t)+n/2]；
If not, then the mobile terminal can be switched to the normal mode,
then to
The following operations are carried out:
(7.2.3) obtaining the jth peak value A j (t) Peak waveform data DataPeek j (t)，j＝2,3,…,(last-1)；
If f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
dataPeek j (t)＝dataP(t)[f j (t)-n/2,f j (t)+n/2]；
if f is j (t)+n/2＜f j+1 (t)&&f j (t)-n/2＜f j-1 (t), then:
then to
The following operations are carried out:
if f is j (t)+n/2＞f j+1 (t)&&f j (t)-n/2＞f j-1 (t), then:
then to
The following operations are carried out:
if none of the above conditions are met:
first to each other
The following operations are carried out:
then to
The following operations are carried out:
(8) Loading labels on all peak waveform data to construct an input data set V of the neural network;
wherein y is a tensor with the size of k × 1, the value of the position corresponding to the power supply number in the tensor of k × 1 is 1, and the values of other positions are 0;
(9) Building a neural network model;
the first layer is an input layer of the neural network, and the data dimension of the input layer is n x 1;
the second layer is a one-dimensional convolution layer, the number of convolution kernels is 128, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and average pooling are used for processing data simultaneously, the size of the pooled kernels is 3 x 1, and the step length is 1;
the third layer is a one-dimensional convolution layer, the number of convolution kernels is 64, the size is 3 x 1, the step length is 1, the activation function is a relu function, batch regularization and maximum pooling are used for processing at the same time, the size of the pooled kernels is 3 x 1, and the step length is 1;
the fourth layer is a one-dimensional convolution layer, the number of convolution kernels is 32, the size is 3 x 1, the step length is 1, and the activation function is a relu function;
the fifth layer is a flattening layer, and the convolution result is expanded into one-dimensional data;
the sixth layer is a fully-connected output layer, the activation function is softmax, and the output dimension is a probability value tensor of k × 1;
(10) Training a neural network model;
(10.1) selecting M groups of peak waveform data in batches from the input data set V and inputting the M groups of peak waveform data into the neural network model, thereby predicting M groups of tensor values of probability of power supply labels
u＝1,2,…,M；
Wherein,
the ith output of the k outputs representing the passage of the u-th set of peak waveform data through the neural network,
the u-th set of peak waveform data is passed through the s-th output of the k outputs of the neural network,
representing a probability value that the u-th group of peak waveform data belongs to the l-th power supply;
(10.2) then calculating the loss function value C of the neural network by the following formula;
wherein,
or
Indicates the category of the u-th group of peak waveform datal belongs to the first power source,
the method comprises the steps that the class l of the u-th group of peak waveform data does not belong to the ith power supply;
(10.3) repeating the steps (10.1) and (10.2), and continuing to train the neural network model until the neural network model converges to obtain the trained neural network model;
(11) Identifying the current sharing state of the redundant power supply;
(11.1) acquiring a total ripple single-sided power spectrum under the current load R (t), and then searching for a frequency f corresponding to the condition meeting the peak value on the single-sided power spectrum of the total ripple according to the method in the step (4) j (t)；
(11.2) acquiring peak waveform data dataPeek with length n on the unilateral power spectrum of the total ripple wave according to the method in the step (7.2) j (t)；
(11.3) mixing j (t) substituting into the fitted power curve L (f, P) to obtain the corresponding power P j (t)；
(11.4) mixing DataPeek j (t) inputting the data to the trained neural network model, thereby predicting the probability value tensor of the power label
(11.5) according to
Calculating a corresponding power supply number;
(11.6) Number according to Power supply j And corresponding output power P j And (t) obtaining the ratio of the current of each power supply, thereby identifying whether each power supply reaches a current equalizing state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110472729.9A CN113189513B (en) | 2021-04-29 | 2021-04-29 | Ripple-based redundant power supply current sharing state identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110472729.9A CN113189513B (en) | 2021-04-29 | 2021-04-29 | Ripple-based redundant power supply current sharing state identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113189513A CN113189513A (en) | 2021-07-30 |
| CN113189513B true CN113189513B (en) | 2022-10-21 |
Family
ID=76980356
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110472729.9A Active CN113189513B (en) | 2021-04-29 | 2021-04-29 | Ripple-based redundant power supply current sharing state identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113189513B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113759271B (en) * | 2021-11-10 | 2022-01-25 | 电子科技大学 | Redundant power supply current sharing state identification method based on frequency spectrum and LSTM network |
| CN114039346A (en) * | 2021-11-19 | 2022-02-11 | 四川启睿克科技有限公司 | Voltage regulation and current sharing method based on neural network |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1273922A1 (en) * | 2001-07-06 | 2003-01-08 | Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. | Methods and devices for measuring electrical currents |
| CN101672692A (en) * | 2009-10-14 | 2010-03-17 | 大连理工大学 | Tuning fork resonance frequency quick measuring method based on virtual instrument |
| CN101802547A (en) * | 2007-07-27 | 2010-08-11 | 实耐宝公司 | Wheel alignment head and system with advanced power management |
| CN103066620A (en) * | 2012-12-24 | 2013-04-24 | 中国电力科学研究院 | Design method of automatic generation control model under intermittent energy grid-connection |
| CN106887840A (en) * | 2017-03-23 | 2017-06-23 | 西安电子科技大学 | A kind of microgrid control system |
| CN108566253A (en) * | 2018-02-12 | 2018-09-21 | 北京邮电大学 | It is a kind of based on the signal recognition method extracted to power spectrum signal fit characteristic |
| CN109188072A (en) * | 2018-07-13 | 2019-01-11 | 中国船舶重工集团公司第七〇九研究所 | A kind of power detection device, radar system |
| JP2020150702A (en) * | 2019-03-14 | 2020-09-17 | ダイキン工業株式会社 | Direct type power conversion device |
| WO2020236726A1 (en) * | 2019-05-18 | 2020-11-26 | Amber Solutions, Inc. | Intelligent circuit breakers |
| CN112306035A (en) * | 2019-08-02 | 2021-02-02 | 中国石油化工股份有限公司 | Diagnostic system for operation fault of chemical process |
| WO2021068454A1 (en) * | 2019-10-12 | 2021-04-15 | 联合微电子中心有限责任公司 | Method for identifying energy of micro-energy device on basis of bp neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10277128B2 (en) * | 2015-06-29 | 2019-04-30 | Semiconductor Components Industries, Llc | Switch control circuit and buck converter comprising the same |
| US10775440B2 (en) * | 2016-08-28 | 2020-09-15 | The Board Of Trustees Of The University Of Alabama | Methods, apparatuses, and systems for measuring impedance spectrum, power spectrum, or spectral density using frequency component analysis of power converter voltage and current ripples |
-
2021
- 2021-04-29 CN CN202110472729.9A patent/CN113189513B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1273922A1 (en) * | 2001-07-06 | 2003-01-08 | Max-Planck-Gesellschaft zur Förderung der Wissenschaften e.V. | Methods and devices for measuring electrical currents |
| CN101802547A (en) * | 2007-07-27 | 2010-08-11 | 实耐宝公司 | Wheel alignment head and system with advanced power management |
| CN101672692A (en) * | 2009-10-14 | 2010-03-17 | 大连理工大学 | Tuning fork resonance frequency quick measuring method based on virtual instrument |
| CN103066620A (en) * | 2012-12-24 | 2013-04-24 | 中国电力科学研究院 | Design method of automatic generation control model under intermittent energy grid-connection |
| CN106887840A (en) * | 2017-03-23 | 2017-06-23 | 西安电子科技大学 | A kind of microgrid control system |
| CN108566253A (en) * | 2018-02-12 | 2018-09-21 | 北京邮电大学 | It is a kind of based on the signal recognition method extracted to power spectrum signal fit characteristic |
| CN109188072A (en) * | 2018-07-13 | 2019-01-11 | 中国船舶重工集团公司第七〇九研究所 | A kind of power detection device, radar system |
| JP2020150702A (en) * | 2019-03-14 | 2020-09-17 | ダイキン工業株式会社 | Direct type power conversion device |
| WO2020236726A1 (en) * | 2019-05-18 | 2020-11-26 | Amber Solutions, Inc. | Intelligent circuit breakers |
| CN112306035A (en) * | 2019-08-02 | 2021-02-02 | 中国石油化工股份有限公司 | Diagnostic system for operation fault of chemical process |
| WO2021068454A1 (en) * | 2019-10-12 | 2021-04-15 | 联合微电子中心有限责任公司 | Method for identifying energy of micro-energy device on basis of bp neural network |
Non-Patent Citations (2)
| Title |
|---|
| Fractional Fourier transform pre-processing for neural networks and its application to object recognition;BillurBarshan 等;《Neural Networks》;20020131;第15卷(第1期);131-140 * |
| 动力电池均衡系统的优化设计;李莉;《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》;20140115;C042-696 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113189513A (en) | 2021-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113189513B (en) | Ripple-based redundant power supply current sharing state identification method | |
| CN113159077B (en) | Three-phase inverter fault identification method based on hybrid convolutional neural network | |
| CN109753762B (en) | Power distribution network two-stage network topology identification method and device based on category correction | |
| CN110147827B (en) | Fault prediction method based on IAALO-SVM and similarity measurement | |
| CN110045317B (en) | Mutual inductor metering error online detection method and system | |
| CN110672905A (en) | CNN-based self-supervision voltage sag source identification method | |
| CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
| CN109507625A (en) | The automatic calibrating method and terminal device of battery simulator | |
| CN108287327A (en) | Metering automation terminal fault diagnostic method based on Bayes's classification | |
| CN112305379A (en) | Mode identification method and system for GIS insulation defect | |
| CN110751213B (en) | Method for identifying and supplementing abnormal wind speed data of wind measuring tower | |
| CN114330486A (en) | Power system bad data identification method based on improved Wasserstein GAN | |
| CN115685046A (en) | Mutual inductor metering abnormity identification method, device, equipment and storage medium | |
| CN114583767B (en) | Data-driven wind power plant frequency modulation response characteristic modeling method and system | |
| CN117972537A (en) | Voltage transformer metering state evaluation method and system based on wide area measurement | |
| CN110766248A (en) | Workshop human factor reliability evaluation method based on SHEL and interval intuition fuzzy evaluation | |
| CN113033077A (en) | Direct-current transmission line fault distance measurement method based on neural network algorithm | |
| CN115308538B (en) | Power distribution network fault positioning method, system, terminal and storage medium based on YOLOV5 | |
| CN108805204B (en) | Electric energy quality disturbance analysis device based on deep neural network and use method thereof | |
| CN115377999B (en) | Subsynchronous oscillation identification method based on broadband measurement data | |
| CN108646145B (en) | Method and system for positioning flashover tower of power transmission line | |
| CN110908365A (en) | Unmanned aerial vehicle sensor fault diagnosis method and system and readable storage medium | |
| CN115169426B (en) | Anomaly detection method and system based on similarity learning fusion model | |
| CN115825648A (en) | Extra-high voltage direct current transmission line fault identification method based on multi-decomposition and parallel residual error network | |
| CN115293090A (en) | Multi-harmonic source responsibility quantification method based on reconstruction data processing |
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 |