The content of the invention
In view of the deficienciess of the prior art, the fault-signal that the present invention is provided in a kind of micro-capacitance sensor failure diagnostic process is special
Value indicative extracting method.
It is the technical scheme is that such:
A kind of fault-signal Eigenvalue Extraction Method in micro-capacitance sensor failure diagnostic process, comprises the following steps:
Step 1:Judge fault-signal from the micro-capacitance sensor voltage of collection or current signal;
For each micro-capacitance sensor voltage or the sampled data in current signal cycle, the sampled data is calculated most
Big value XimaxWith minimum value Ximin;According to XimaxAnd XiminJudge whether the sampled data is symmetrical, and then judge the hits
Whether it is fault-signal according to corresponding micro-capacitance sensor voltage or current signal, method is:If Ximax=-Ximin, then the hits
According to symmetrical, show that the micro-capacitance sensor voltage corresponding to the sampled data or current signal are normal signal;If Ximax≠-Ximin, then
The sampled data is asymmetric, shows that the micro-capacitance sensor voltage corresponding to the sampled data or current signal are fault-signal;
Step 2:The fault-signal that step 1 obtains is reconstructed using extreme value symmetrization processing method, makes its symmetrical;
For the sampled data in each fault-signal cycle:Maximum value is calculated first, and the maximum is absolute
It is worth corresponding fault-signal element to be denoted as;Then fault-signal is reconstructed using extreme value symmetrization processing method, side
Method is:
a):Assuming that there is N number of point in each fault-signal cycle, the corresponding fault-signal element of the N number of point is denoted asFault-signal is divided into two parts, is respectively:With
b):According toPosition in fault-signal, is reconstructed fault-signal, and then obtains failure reconfiguration letter
Number;
IfPositioned at XhIn, utilize fault-signal reconstruction formula Xq=-XhFault-signal is reconstructed, obtains event
Hinder reconstruction signal X 'i=Xq+Xh;IfPositioned at XqIn, then utilize fault-signal reconstruction formula Xh=-XqTo fault-signal into
Row reconstruct, obtains failure reconfiguration signal X 'i=Xq+Xh;
Step 3:Failure reconfiguration signal is normalized;
Wherein X 'inewFor the failure reconfiguration signal after normalized;(Min, Max) is normalization amplitude section;
Step 4:Using equation below, the amplitude size of the failure reconfiguration signal after normalized is adjusted to original event
Hinder the amplitude size of signal, obtain the fault-signal after normalized;
Wherein, PiFor the amplitude proportion of the failure reconfiguration signal after primary fault signal and normalized;P is amplitude base
Quasi- value;
Step 5:Preliminary extraction fault-signal characteristic value, including:The main band energy value of fault-signal, fault-signal are average
Rate of change, fault-signal amplitude average value and fault-signal root-mean-square value;
Step 5-1:The main frequency range energy of fault-signal is obtained using the multiresolution analysis method based on wavelet transform
Value;
1) using the multiresolution analysis method based on wavelet transform to XinewThe decomposition of multilayer is carried out, obtains coefficient
Aj、Dj;Wherein j is the number of plies;AjFor fault-signal XinewIn the wavelet coefficient of the approximate part of jth layer;DjFor fault-signal Xinew
In the wavelet coefficient of the detail section of jth layer;
2) to every layer coefficients Aj、DjIt is reconstructed to obtain the corresponding signal of different frequency range, and to the reconstruction signal of different frequency range
Its energy value is calculated, and chooses a characteristic value of the energy value as fault-signal of main frequency range, the i.e. main frequency of fault-signal
Section energy value;
Step 5-2:Obtain fault-signal mean change speed;Fault-signal mean change rate representation is in some cycles
The average speed of fault-signal change;Fault-signal mean change speed | s ' |avCalculating formula is:
In formula, f0For the frequency of fault-signal;| s ' (j) | for the absolute value of fault-signal rate of change, NnFor fault-signal
Element number;
Step 5-3:Obtain fault-signal amplitude average value and fault-signal root-mean-square value;
Fault-signal amplitude average value savRepresent the amplitude size of this periodic signal, calculating formula is:
Fault-signal root-mean-square value srmsRepresent that fault-signal deviates the degree of its average value, calculating formula is:
In formula, s (i) is fault-signal amplitude, NnFor the element number of fault-signal;
Step 6:Processing is optimized to the fault-signal characteristic value tentatively extracted in step 5, exports final failure letter
Number characteristic value is to micro-capacitance sensor fault verification process;
Step 6-1:Fault-signal characteristic value is gathered in order, and characteristic value is stored into characteristic value memory successively;It is described
The capacity of characteristic value memory is Lw;
Step 6-2:Transition threshold value J is setm, judging characteristic value X (n) is normal condition or transition state, and method is:Root
According to X (n)-X (n-1) and transition threshold value JmRelative size relation, judge n-th of characteristic value state:X if (n)-X (n-1) is small
In transition threshold value Jm, then this feature value is normal condition, performs step 6-3;Otherwise, this feature value is transition state Sta, empty
Characteristic value memory, and output characteristic value X (n) is to micro-capacitance sensor fault verification process;
Step 6-3:Left window average value processing is directly carried out to this feature value, calculates being averaged for primitive character value X (n)
Value Xnew(n) characteristic value after i.e. preliminary optimization, and by Xnew(n) it is stored into characteristic value average store, characteristic value is average
The capacity of value memory is denoted as LT;
Left window statistical average handles formula:
Wherein JwjwIt is characterized value number;
Step 6-4:To the characteristic value X after preliminary optimizationnew(n) further optimization is done, exports final fault-signal feature
It is worth to micro-capacitance sensor fault verification process;
Step 6-4-1:Find the steady state value of characteristic value;Method is:To characteristic value in characteristic value average store
Average value subtracts each other successively according to order from back to front, obtains NsA difference, i.e. Xnew(n)-Xnew(n-1), Xnew(n-1)-Xnew
(n-2) ..., Xnew(n-Ns+1)-Xnew(n-Ns), if all differences are respectively less than stable threshold Sth, then by Xnew(n) it is recorded as spy
The present steady state value of value indicative, if its again with all existing history steady state value VstDifference, then by present steady state
Value Xnew(n) storage is into steady state value memory;
Step 6-4-2:By the characteristic value average value in characteristic value average store according to order dot interlace from back to front
Subtract successively with the steady state value in its immediate steady state value memory, obtain NcA difference, i.e. Xnew(n)-Vst
(i), Xnew(n-2)-Vst(i),…,Xnew(n-2(Nc-1))-Vst(i), if all differences are respectively less than threshold value Sc, and these differences
It is gradually reduced, then think this characteristic value Xnew(n) a certain stable state is being tended to, then to the characteristic value X after preliminary optimizationnew
(n) further optimization is done, exports final fault-signal characteristic value to micro-capacitance sensor fault verification process;
To the characteristic value X after preliminary optimizationnew(n) doing the calculation formula further optimized is:
X'new(n)=Xnew(n)+k(Sst(i)-Xnew(n))
In formula, k is regulation coefficient;Sst(i) it is a certain in stable condition value;
Beneficial effects of the present invention:The present invention bases oneself upon the change of signal characteristic before and after failure, is handled by extreme value symmetrization,
Preliminary failure judgement signal and normal signal, suppression signal fluctuation, retention fault signal characteristic, normalized fault-signal,
Pending fault-signal is idealized, is then returned from multi-angle using extraction the methods of being based on discrete wavelet multiresolution analysis
The key character value of fault-signal after one change, finally using self-adaptive features value dynamic optimization method, to reach automated, multi-level
Optimization processing micro-capacitance sensor fault eigenvalue purpose, for follow-up micro-capacitance sensor fault diagnosis provide stabilization fault eigenvalue.
Embodiment
Embodiments of the present invention are described in further detail below in conjunction with the accompanying drawings.
Fault-signal in micro-capacitance sensor is usually randomness non-stationary classification, fault-signal be characterized by its relative to
In the abrupt information of normal signal, the Mutational part of signal just becomes the important transient state characteristic value of signal, is one instantaneous
Knots modification, if failure cannot improve for the moment, continuous abrupt information occurs in this failure transient characteristic value.Mutation letter
Breath represents to describe by the general performance of the singular point of the wavelet transformation of signal on multiple dimensioned.Therefore, research event
The abrupt information of barrier signal can be obtained by the singularity of signal Analysis wavelet transformation, this also just becomes one very heavy
Want and useful signal processing means.
In micro-capacitance sensor fault diagnosis, the mutation singular point for extracting fault-signal is exactly the main information for having grasped failure.
On the other hand, if the fault-signal characteristic information extracted can reconstruct, the characteristic value extracted can be used for data again
Compression.Therefore, there is the abrupt information that signal is extracted in the fault diagnosis of micro-capacitance sensor, carrying out the Singularity Analysis of wavelet transformation is
Significantly.
Present embodiment describes micro- electricity of present embodiment in detail by taking the microgrid inverter switch fault shown in Fig. 1 as an example
Fault-signal Eigenvalue Extraction Method in net failure diagnostic process.Fig. 1 is the micro-capacitance sensor structure diagram based on VF controls, is wrapped
Include equivalent distributed generation resource, battery storage equipment, VF controllers, PWM drive modules, inverter, LC wave filters, transmission line,
Bus and different loads.Equivalent Distributed formula power supply is the offer energy in micro-capacitance sensor, and equivalent process is DC power supply;VF is controlled
Device processed is used for island mode control voltage and frequency.PWM drivings are for driving inverter.Inverter is will be acquired straight
Circulation is changed to the three-phase alternating current of needs;Transmission line is different in size, can substantially be equivalent to the series connection of resistance and inductance;Bus is
Connect used in other networks, present embodiment is assumed to be island mode, without considering major network.In addition different varying duties is connected to
In bus.Present embodiment will be controlled by using VF, the treatment effect of fault-signal in the case of island mode, to illustrate this
The applicability of inventive method is stronger, more closing to reality.
Fault-signal Eigenvalue Extraction Method in the micro-capacitance sensor failure diagnostic process of present embodiment, as shown in Fig. 2, tool
Body includes the following steps:
Step 1:The micro-capacitance sensor voltage signal gathered from Fig. 1, and therefrom judge fault-signal;
For the sampled data in each micro-capacitance sensor voltage cycle, the maximum X of the sampled data is calculatedimaxMost
Small value Ximin;According to XimaxAnd XiminJudge whether the sampled data is symmetrical, and then judge micro- corresponding to the sampled data
Whether network voltage is fault-signal, and method is:If Ximax=-Ximin, then the sampled data is symmetrical, shows the sampled data institute
Corresponding micro-capacitance sensor voltage is normal signal;If Ximax≠-Ximin, then the sampled data is asymmetric, shows that sampled data institute is right
The micro-capacitance sensor voltage answered is fault-signal, as shown in Figure 3;
Step 2:The fault-signal that step 1 obtains is reconstructed using extreme value symmetrization processing method, makes its symmetrical;
For the sampled data in each fault-signal cycle:Maximum value is calculated first, and the maximum is absolute
It is worth corresponding fault-signal element to be denoted as;Then fault-signal is reconstructed using extreme value symmetrization processing method, such as
Shown in Fig. 4, method is:
a):Assuming that there is N number of point in each fault-signal cycle, the corresponding fault-signal element of the N number of point is denoted asFault-signal is divided into two parts, is respectively:With
b):According toPosition in fault-signal, is reconstructed fault-signal, and then obtains failure reconfiguration letter
Number;
IfPositioned at XhIn, utilize fault-signal reconstruction formula Xq=-XhFault-signal is reconstructed, obtains event
Hinder reconstruction signal X 'i=Xq+Xh;IfPositioned at XqIn, then utilize fault-signal reconstruction formula Xh=-XqTo fault-signal into
Row reconstruct, obtains failure reconfiguration signal X 'i=Xq+Xh;
Step 3:Failure reconfiguration signal is normalized;
Wherein X 'inewFor the failure reconfiguration signal after normalized;(Min, Max) is normalization amplitude section;
Step 4:Using equation below, the amplitude size of the failure reconfiguration signal after normalized is adjusted to original event
Hinder the amplitude size of signal, obtain the fault-signal after normalized;
Wherein, PiFor the amplitude proportion of the failure reconfiguration signal after primary fault signal and normalized;P is amplitude base
Quasi- value;
Fig. 3 and Fig. 5 are contrasted and understood, signal after extreme value symmetrization is handled, protected by main fault-signal feature
Stay, be suppressed by the fluctuation signal of load change and system self-adaption adjustment.And after treatment, fault-signal is more preferable,
It is more advantageous to extraction characteristic value.
Step 5:Preliminary extraction fault-signal characteristic value, including:The main band energy value of fault-signal, fault-signal are average
Rate of change, fault-signal amplitude average value and fault-signal root-mean-square value;
Step 5-1:The main frequency range energy of fault-signal is obtained using the multiresolution analysis method based on wavelet transform
Value, detailed process such as Fig. 5 (a), shown in 5 (b) and 5 (c);
1) using the multiresolution analysis method based on wavelet transform to Xinew11 layers of decomposition is carried out, such as Fig. 5 (a)
It is shown, obtain coefficient Aj、Dj;Wherein j is the number of plies;AjFor fault-signal XinewIn the wavelet coefficient of the approximate part of jth layer;DjFor
Fault-signal XinewIn the wavelet coefficient of the detail section of jth layer;
Wherein:
In formula, j, k and n are integers, and j is the wavelet decomposition number of plies, Aj,kFor wavelet decomposition approximation coefficient, Dj,kFor wavelet decomposition
Detail coefficients, φj,k(t) it is extension function, ψj,k(t) it is wavelet function.The generating function of wavelet decomposition is chosen in present embodiment
‘db3’。
2) to every layer coefficients Aj、DjIt is reconstructed to obtain the corresponding signal of 11 frequency ranges shown in Fig. 5 (b), and to not
Reconstruction signal with frequency range calculates its energy value, and calculation formula is:
Its medium size EDjRepresent jth layer energy value, DjRepresent jth layer reconstruction signal, NjRepresent the element of jth layer reconstruction signal
Number, μjRepresent the average of jth layer signal.
Choose a characteristic value of the energy value as fault-signal of main frequency range, the i.e. main band energy of fault-signal
Value;By calculating the energy value of reconstruction signal, fault-signal X is obtainedinew11 frequency band energy values, as shown in Fig. 5 (c)
Energy ingredient figure, pass through analysis understand fault-signal XinewThe energy of middle 80%-90% concentrates on certain several frequency band, in order to
Main energy variation situation can be caught again by reducing Characteristic Number, choose characteristic value of the energy value as signal of main frequency range.
Most common failure is open fault, short trouble in microgrid inverter failure.According to the symmetrical of inverter topology
Property understand, a pair need to only be studied by studying its switch fault, in order to save space, using S1 and S2 failures to illustrate object.According to
According to different frequency range energy ingredient figure, such as Fig. 6, understands through analysis contrast, no matter signal be it is normal, or switch short or short circuit therefore
Hinder, the energy of 80%-90% always concentrates on the 9th layer and the 10th layer in its signal.9th layer and the 10th layer of energy value is distinguished
It is defined as ED9 and ED10.Therefore, come out ED9, ED10 as the characteristics extraction of signal, be rationally accurate.Therefore will
The characteristic value of ED9, ED10 are as the main band energy value of microgrid inverter fault diagnosis fault-signal in present embodiment
Characteristic value, it not only improves the accuracy of fault diagnosis, has caught most important component, and reduce the characteristic value of fault diagnosis
Input quantity, reduce calculation amount.
Step 5-2:Obtain fault-signal mean change speed;Signal averaging rate of change is represented in some cycles internal fault
The average speed of signal intensity;Fault-signal mean change speed | s ' |avCalculating formula is:
In formula, f0For the frequency of fault-signal;| s ' (j) | for the absolute value of fault-signal rate of change, NnFor fault-signal
Element number;
Step 5-3:Obtain fault-signal amplitude average value and fault-signal root-mean-square value;
Fault-signal amplitude average value savRepresent the amplitude size of this periodic signal, calculating formula is:
Fault-signal root-mean-square value srmsRepresent that fault-signal deviates the degree of its average value, calculating formula is:
In formula, s (i) is fault-signal amplitude, NnFor the element number of fault-signal;
Step 6:Self-adaptive features value dynamic optimization, such as Fig. 7 are carried out to the fault-signal characteristic value tentatively extracted in step 5
Show, export final fault-signal characteristic value to micro-capacitance sensor fault verification process;
Step 6-1:Characteristic value, is stored into characteristic value memory by gathered data characteristic value in order successively;The feature
The capacity for being worth memory is Lw;
Step 6-2:Transition threshold value J is setm, judging characteristic value X (n) is normal condition or transition state, and method is:Root
According to X (n)-X (n-1) and transition threshold value JmRelative size relation, judge n-th of characteristic value state:X if (n)-X (n-1) is small
In transition threshold value Jm, then this feature value is normal condition, performs step 6-3;Otherwise, this feature value is transition state Sta, empty
Characteristic value memory, and output characteristic value X (n) is to micro-capacitance sensor fault verification process;
Step 6-3:Left window average value processing is directly carried out to this feature value, calculates being averaged for primitive character value X (n)
Value Xnew(n) characteristic value after i.e. preliminary optimization, and by Xnew(n) it is stored into characteristic value average store, characteristic value is average
The capacity of value memory is denoted as LT;
Left window statistical average handles formula:
Wherein JwjwIt is characterized value number;
Step 6-4:To the characteristic value X after preliminary optimizationnew(n) further optimization is done, exports final fault-signal feature
It is worth to micro-capacitance sensor fault verification process;
Step 6-4-1:Find the steady state value of characteristic value;Method is:To characteristic value in characteristic value average store
Average value subtracts each other successively according to order from back to front, obtains NsA difference, i.e. Xnew(n)-Xnew(n-1), Xnew(n-1)-Xnew
(n-2) ..., Xnew(n-Ns+1)-Xnew(n-Ns), if all differences are respectively less than stable threshold Sth, then by Xnew(n) it is recorded as spy
The present steady state value of value indicative, if its again with all existing history steady state value VstDifference, then by present steady state
Value Xnew(n) storage is into steady state value memory;
Step 6-4-2:By the characteristic value average value in characteristic value average store according to order dot interlace from back to front
Subtract successively with the steady state value in its immediate steady state value memory, obtain NcA difference, i.e. Xnew(n)-Vst
(i), Xnew(n-2)-Vst(i),…,Xnew(n-2(Nc-1))-Vst(i), if all differences are respectively less than threshold value Sc, and these differences
It is gradually reduced, then think this characteristic value Xnew(n) a certain stable state is being tended to, then to the characteristic value X after preliminary optimizationnew
(n) further optimization is done, exports final fault-signal characteristic value to micro-capacitance sensor fault verification process;
To the characteristic value X after preliminary optimizationnew(n) doing the calculation formula further optimized is:
X'new(n)=Xnew(n)+k(Sst(i)-Xnew(n))
In formula, k is regulation coefficient;Sst(i) it is a certain in stable condition value;
In order to illustrate effect, with VaEffect carries out comparative illustration before and after the optimization of voltage changing rate characteristic value, such as Fig. 8 institutes
Show.When Fig. 8 (a) occurs for S1 open circuit faults, VaThe change of voltage signal.Fig. 8 (b) is corresponding VaVoltage signal rate of change
The front and rear effect contrast figure of characteristic value optimization.It is obvious to can see, larger fluctuation can be produced to characteristic value, lead to when the load
Real-time on-line optimization processing is crossed, system can be with the same state characteristic value of automatic identification, and according to circumstances makes different schemes
Processing, realizes stabilization of the characteristic value under same state.When system saltus step, such as figure latter half, normal condition is returned in saltus step, because
Exist for system and be self-regulated, the fluctuation of certain period of time occurs under normal circumstances, then tends to some value, and in a certain range
Fluctuation, but be fast and accurately identified by this optimization method, characteristic value, and Long-term change trend is accelerated, realize preferable
Correct effect of optimization.Characteristic value being stabilized under different conditions is realized, reduces load change and System Small Disturbance band
The characteristic value interference problem come.And this module has on-line study ability, when inputting certain characteristic value, start automatic
Study, improves the resolving ability of itself.
The fault identification of micro-capacitance sensor is carried out in present embodiment using artificial neural network, utilizes micro- electricity of present embodiment
The fault-signal characteristic value of fault-signal Eigenvalue Extraction Method extraction in net failure diagnostic process, as artificial neural network
Input, the artificial neural network include 15 input, 12 output.Wherein, the input of 15 artificial neural networks includes:Bag
ED9, ED10 are included, | s ' |av, sav, srms, the phase of micro-capacitance sensor three-phase voltage;12 outputs represent microgrid inverter switch event
13 kinds of fault types of barrier, as shown in Figure 9.By extreme value symmetrical treatment, characteristics extraction and optimization, artificial neural network it is defeated
Enter quantity to greatly reduce, so as to reduce the difficulty of fault diagnosis, and improve the application range of artificial neural network.
The sample frequency of this implementation system is 50KHZ, randomly selects 390 groups of data in 13 kinds of fault types as experiment
Sample.Wherein, 260 groups of experiment samples are the training sample of artificial neural network, and remaining 130 groups of experiment samples are artificial neuron
The test sample of network.Test result is as shown in table 1.From table 1 it follows that the accuracy of fault identification reaches 100%.
Table 1
In conclusion the fault-signal Eigenvalue Extraction Method in the micro-capacitance sensor failure diagnostic process of present embodiment, right
The fault identification of micro-capacitance sensor has extensive practical significance, is particularly used for micro-capacitance sensor fault identification to artificial neural network, has very
Big lifting.The input quantity of artificial neural network is greatly reduced, thus greatly reduces the calculation amount of artificial neural network,
And greatly improve the accuracy of artificial neural network identification micro-capacitance sensor failure and rapidity.