Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention discloses a method for detecting foreign matter defects in GIS equipment. As shown in fig. 1, the method comprises the steps of:
step S101: and training a foreign matter defect classification model by adopting a training sample of the sound signal in the GIS equipment.
It should be understood that each training sample corresponds to a standard class of foreign object defects.
The foreign object defect classification model is an Extended continuous adaptive fuzzy inference system (ESAFIS). The ESAFIS is a sequence learning algorithm, and the ESAFIS establishes a rule base by means of a functional equivalence relation between a Radial Basis Function (RBF) neural network and a Fuzzy Inference System (FIS), and the structure changes adaptively along with the input of data. ESAFIS belongs to a typical feed-forward neural network.
As shown in fig. 2, the extended continuous adaptive fuzzy inference system includes an input layer, a fuzzy inference layer, a specification layer and an output layer, which are sequentially arranged. The input layer is used for inputting at least one-dimensional vector to be classified each time, specifically to an application scene for detecting the type of the foreign object defect, the vector to be classified is a sound signal, and the dimension of the sound signal is one-dimensional. The fuzzy layer is closely connected with the fuzzy inference layer, and each fuzzy rule has a membership function. The node number of the fuzzy layer changes along with the addition and deletion of the fuzzy rule. Each node of the fuzzy inference layer represents a fuzzy rule. The ESAFIS fuzzy rule evolution is based on a distance criterion and a correction influence standard, and nodes are adaptively added and deleted along with the input of data. The standard layer is used for normalizing the output of the fuzzy inference layer, and the number of the nodes is the same as that of the fuzzy inference layer. The output layer is used for outputting the classification result, and the number of the nodes is the same as the number of the classified categories.
The fuzzy rule with sequence number k of the extended continuous adaptive fuzzy inference system can be expressed as: if (x)
1Is A
ik),……,(x
iIs A
ik),……(
Is that
) Then (1)
Is a
k1),……,(
Is a
kj),……,(
Is that
). Wherein the vector is input
A
ikRepresenting the membership value of the ith component of the input vector to the fuzzy rule with the sequence number k, and outputting the vector
a
kjThe jth back-piece parameter of the fuzzy rule with the sequence number k is shown. k is 1, 2, … …, N
h。i=1,2,……,N
x。j=1,2,……,N
y。N
hIndicating the number of fuzzy rules. N is a radical of
xAnd N
yBoth represent dimensions. For the application scenario of the classification of the foreign object defect inside the GIS device of the embodiment of the present invention, N
x=1,N
y=1。
The excitation function of the fuzzy rule with the sequence number of k of the extended continuous adaptive fuzzy inference system is a Gaussian function, and can be calculated by using a general formula of the excitation function:
wherein R isk(x) When a vector x is input, the excitation function of the fuzzy rule with the sequence number k is obtained through calculation. SigmakRadius of excitation function, mu, of fuzzy rule with index kkRepresenting the center of the excitation function of the fuzzy rule with index k. For the application scenario of the classification of the foreign object defect inside the GIS device in the embodiment of the present invention, x is the input sound signal.
The output vector of the extended continuous adaptive fuzzy inference system can be calculated by the following output vector general formula:
wherein, a
kAnd expressing a weight matrix from the fuzzy rule with the sequence number k to the specification layer, namely a back-part parameter matrix. For the first order Takagi-Sugeno (TS) fuzzy model,
x
eextended by x and 1, i.e. x
e=[1,x]
T。q
kIs a parameter matrix of the fuzzy rule with the sequence number k. The general formula of the parameter matrix is as follows:
as can be seen from the above description of the extended continuous adaptive fuzzy inference system, the output vector is affected by the input vector, the fuzzy rule and the parameters of the back-part. The parameters of the back-part are further subjected to a parameter matrix qkThe fuzzy rule itself is affected by muk、σkEtc. of the parameters. Therefore, the fuzzy rules of the extended continuous adaptive fuzzy inference system are not fixed, but dynamically adjusted in the learning process, so as to add the fuzzy rules, delete the fuzzy rules or update the corresponding parameters.
Specifically, step S101 includes the following steps:
1. and inputting each training sample in turn according to the sequence number of the training sample.
Every time a training sample is input, a fuzzy rule is added to the assumed foreign object defect classification model.
It should be appreciated that the training samples described in the embodiments of the present invention are novel. Specifically, if the training samples satisfy psi<EsThen the training sample has novelty.
Where ψ represents the spherical potential energy of the training sample in the high-dimensional feature space. The high-dimensional feature space appears spherical. The embodiment of the invention describes the high-dimensional feature space by using the front piece parameters of the Gaussian rule, including the central point mu and the width sigma.
T denotes the dimension of the high-dimensional feature space, phi (x, mu)
t) And representing the mapping function of the high-dimensional feature space corresponding to the training sample. Mu.s
tRepresenting the center of the t-th mapping function. It should be understood that the mapping function of embodiments of the present invention is a gaussian function. E
sThe novelty threshold is expressed and can be preset empirically. Only if the training sample meets the above-mentioned novelty condition, the training sample is inputted into the extended continuous adaptive fuzzy inference system.
2. If the sequence number of the currently input training sample is 1, the fuzzy rule added correspondingly to the currently input training sample is reserved.
It should be appreciated that the extended continuous adaptive fuzzy inference system has an initial number of fuzzy rules of 0 before inputting the training samples. Therefore, when the first training sample is input, no judgment is made, and the fuzzy rule added corresponding to the input training sample is reserved.
3. And if the serial number of the currently input training sample is greater than 1, judging whether to keep the fuzzy rule which is correspondingly added to the currently input training sample.
After the first training sample is input, since the extended continuous adaptive fuzzy inference system already has a fuzzy rule, the input of other training samples and the assumption of the added fuzzy rule need to be judged so as to determine whether the fuzzy rule can be added.
Specifically, the steps include the following processes:
(1) and judging whether the added fuzzy rules corresponding to the currently input training sample meet the distance criterion and the influence criterion.
Specifically, the distance criterion is as follows:
||xn-μnr||>en。
xnwhich represents the currently entered training sample with sequence number n. n is>1。μnrRepresenting distance training samples xnThe center of the nearest fuzzy rule. Nearest refers to training sample xnAnd the center of the fuzzy rule is the smallest. e.g. of the typenRepresenting a distance threshold. In particular, en=max{emax×γn,emin}. Wherein e ismaxThe maximum value representing the distance threshold may be preset empirically. e.g. of the typeminThe minimum value representing the distance threshold may be preset empirically. Gamma is a decaying exponential, is a positive number less than 1, and can be preset empirically, so enInitially at a maximum and then decays exponentially until the decay is at a minimum.
Specifically, the impact criteria are as follows:
Emin f(Nh+1)>eg。
wherein E ismin f(Nh+1) represents the impact value of the currently input training sample for the added fuzzy rule. N is a radical ofhIndicating the number of fuzzy rules that exist. e.g. of the typegIndicating that the threshold is increased, may be preset empirically.
Example E of the inventionmin f(Nh+1) is calculated using the formula:
wherein x is
lRepresentation for calculating E
min f(N
h+1) training samples with sequence number l.
Representing the use of training samples x
lThe calculated fuzzy rule (with the sequence number N) which is correspondingly added to the currently input training sample
h+1) excitation function. R
k(x
l) Representing the use of training samples x
lAnd calculating an excitation function of the fuzzy rule with the sequence number k. It should be understood that the sequence numbers of the fuzzy rules are arranged in order of retention. M represents the number of most recently input training samples including the currently input training sample with sequence number n.
With reference to the general formula of the excitation function, Rk(xl) Calculated using the formula:
wherein σkRadius of excitation function, mu, of fuzzy rule with index kkRepresenting the center of the excitation function of the fuzzy rule with index k.
In the same way as above, the first and second,
wherein the content of the first and second substances,
fuzzy rule (with sequence number N) representing corresponding increase of currently input training sample
h+1) radius of the excitation function of the fuzzy rule,
fuzzy rule (with sequence number N) representing corresponding increase of currently input training sample
h+1) center of the excitation function.
(2) If the distance criterion and the influence criterion are met, the fuzzy rules which are correspondingly added to the currently input training samples are reserved.
If both the distance criterion and the influence criterion are met, the added fuzzy rule may be retained, otherwise, the added fuzzy rule is not retained,
4. and if the added fuzzy rule corresponding to the currently input training sample is reserved, setting the parameter of the added fuzzy rule corresponding to the currently input training sample according to the first-order Takagi-Sugeno fuzzy model.
Specifically, the parameters of the fuzzy rule mainly include: parameter matrix, center and radius of the excitation function.
Therefore, this step specifically sets the corresponding parameters by the following procedure:
(1) and setting a parameter matrix of the fuzzy rule which is added corresponding to the currently input training sample by adopting a first equation group.
Referring to the general formula of the parameter matrix above, the first set of equations is as follows:
wherein j is 1, 2, … …, Ny。i=1,2,……,Nx。
Because of the application scenario of the classification of the foreign object defect inside the GIS device of the embodiment of the present invention, N
x=1,N
yWhen the current input training sample corresponds to the added fuzzy rule, the parameter matrix is 1
Thus, the first set of equations is as follows:
(2) and setting the center of the excitation function of the added fuzzy rule corresponding to the currently input training sample by adopting a second equation.
Wherein the second equation is as follows:
(3) and setting the radius of the excitation function of the added fuzzy rule corresponding to the currently input training sample by adopting a third program.
Wherein the third process is as follows:
k represents an overlap factor for controlling the overlap of the added fuzzy rule in the input space, which can be preset empirically.
Therefore, the relevant parameters of the currently input training sample corresponding to the added fuzzy rule can be set through the three equations (sets) described above.
5. And if the fuzzy rule which is added correspondingly to the currently input training sample is not reserved, judging whether to delete the existing fuzzy rule or not after updating the back part parameters of the existing fuzzy rule of the foreign body defect classification model by adopting a least square recursive error method (RLSE).
Specifically, the equation set of the least squares recursive error method is as follows:
wherein, a
nA latter parameter matrix representing a fuzzy rule with a sequence number n. y is
n-1And the standard type of the foreign body defect corresponding to the training sample with the sequence number of n-1 is shown.
The type of the foreign body defect output by the foreign body defect classification model can be calculated by the output vector general formula when the training sample with the sequence number of n-1 is input into the foreign body defect classification model. I denotes an identity matrix. I is a
1×z
1An identity matrix of dimensions. z is a radical of
1And representing the dimensionality of the back-piece parameter of the added fuzzy rule corresponding to the currently input training sample. It should be understood that when the training samples of the current input correspond to the added fuzzy rule, P
nAlso increase in the dimension of (i.e. the
H0The initial value representing the preset is a constant, which can be set empirically. HnRepresenting an input xnThe output vector of the time normalization layer is specifically calculated by adopting the following formula:
see xeGeneral formula (II) of (II), can know that xne=[1,xn]T。Rk(xn) The general calculation of the excitation function is referred to above and will not be described in detail here.
6. If the existing fuzzy rule is deleted, the parameter of the existing fuzzy rule is deleted, and the number of the existing fuzzy rules is updated.
For example, deleting 1 fuzzy rule, then subtracting 1 from the number of existing fuzzy rules.
And finishing the training process after all the training samples are input into the foreign matter defect classification model for training. Through the training process, the fuzzy rule and the back piece parameters of the foreign matter defect classification model can be determined, so that the foreign matter defect classification model can classify the foreign matter defects more accurately.
Step S102: and collecting sound signals inside the GIS equipment at preset time intervals.
The preset time interval may be set empirically. The sound signal collection can be realized by an electronic auscultation technology. The electronic auscultation technology can improve the efficiency and accuracy of data acquisition inside the GIS equipment. In particular, physical signals inside the GIS device, such as ultrasonic signals, vibration signals, etc., may be acquired. The physical signal is sequentially converted into an electric signal, a digital signal and an analog signal through corresponding electric elements, and finally the analog signal is converted into a sound signal. For example, the corresponding sensor may be connected to the GIS device to be tested, and a physical signal generated by a foreign object defect inside the GIS device may be converted into a corresponding electrical signal. During the conversion, conventional amplification processing, filtering processing, and the like may be performed by the respective electric elements.
Step S103: and inputting the sound signal into the trained foreign matter defect classification model, and outputting the type of the foreign matter defect corresponding to the sound signal.
Specifically, the feature values of the sound signals, which may be frequencies, amplitudes, etc., may be extracted and input to the foreign object defect classification model. The feature value of the sound signal is input to the foreign matter defect classification model from the input layer, and the output layer outputs the type of the foreign matter defect corresponding to the sound signal.
In a preferred embodiment of the present invention, the types of the foreign object defect include: 0.5mm spherical particles, 1mm spherical particles, 2mm spherical particles, 1mm linear particles, 2mm linear particles. The accuracy of classification by the foreign matter defect classification model was 98%.
In summary, the method for detecting the foreign object defect inside the GIS device according to the embodiment of the present invention uses the extended continuous adaptive fuzzy inference system with good classification rate obtained by training as the classification model of the foreign object defect, and inputs the sound signal obtained by the electronic auscultation technology into the classification model of the foreign object defect to obtain the type of the foreign object defect, thereby having higher accuracy and efficiency.
The embodiment of the invention also discloses a system for detecting the foreign matter defect in the GIS equipment. As shown in fig. 3, the detection system includes the following modules:
the training module 301 is configured to train a foreign object defect classification model by using a training sample of a sound signal inside the GIS device.
Wherein, each training sample corresponds to a standard type of the foreign body defect.
And the acquisition module 302 is configured to acquire a sound signal inside the GIS device at preset time intervals.
The classification module 303 is configured to output a type of the foreign object defect corresponding to the sound signal after the sound signal is input into the trained foreign object defect classification model.
The foreign matter defect classification model is an extended continuous adaptive fuzzy inference system, the extended continuous adaptive fuzzy inference system comprises an input layer, a fuzzy inference layer, a specification layer and an output layer which are sequentially arranged, and nodes of the fuzzy inference layer represent fuzzy rules.
Preferably, the training module 301 comprises:
and the input submodule is used for sequentially inputting each training sample according to the sequence number of the training sample.
Wherein, when inputting a training sample, the foreign body defect classification model is supposed to be added with a fuzzy rule; the training samples have a novelty.
If the training samples satisfy psi<E
sThen the training sample has novelty. Wherein psi represents the spherical potential energy of the training sample in the high-dimensional feature space,
t denotes the dimension of the high-dimensional feature space, phi (x, mu)
t) A mapping function, mu, representing a high-dimensional feature space corresponding to the training samples
tRepresenting the center of the t-th mapping function, which is a Gaussian function, E
sRepresenting a novelty threshold.
And the retention submodule is used for retaining the fuzzy rule which is correspondingly added to the currently input training sample if the serial number of the currently input training sample is 1.
And the first judgment submodule is used for judging whether to keep the fuzzy rule which is correspondingly added to the currently input training sample if the serial number of the currently input training sample is greater than 1.
And the setting submodule is used for setting parameters of the fuzzy rule which is added to the currently input training sample according to the first-order Takagi-Sugeno fuzzy model if the fuzzy rule which is added to the currently input training sample is reserved.
And the second judgment submodule is used for judging whether to delete the existing fuzzy rule or not after updating the back part parameters of the existing fuzzy rule of the foreign matter defect classification model by adopting a least square recursive error method if the fuzzy rule which is correspondingly added to the currently input training sample is not reserved.
And the deleting submodule is used for deleting the parameters of the existing fuzzy rules and updating the quantity of the existing fuzzy rules if the existing fuzzy rules are deleted.
Preferably, the first judgment sub-module includes:
the first judging unit is used for judging whether the added fuzzy rule corresponding to the currently input training sample meets the distance criterion and the influence criterion.
And the retaining unit is used for retaining the fuzzy rule which is correspondingly added to the currently input training sample if the distance criterion and the influence criterion are met.
Wherein, the distance criterion is | | xn-μnr||>en,xnTraining samples with sequence number n representing the current input, n>1,μnrRepresenting distance training samples xnCenter of the nearest fuzzy rule, enRepresenting a distance threshold. e.g. of the typen=max{emax×γn,eminIn which emaxRepresents the maximum value of the distance threshold, eminRepresents the minimum value of the distance threshold and gamma represents the decay exponent.
Wherein the influence criterion is E
min f(N
h+1)>e
g,N
hIndicating the number of fuzzy rules present, E
min f(N
h+1) represents the influence value of the currently input training sample on the added fuzzy rule, e
gIndicating that the threshold is increased.
Wherein,x
lRepresentation for calculating E
min f(N
h+1) training samples with sequence number l,
representing the use of training samples x
lCalculating an excitation function, R, of the fuzzy rule added to the currently input training sample
k(x
l) Representing the use of training samples x
lAnd calculating an excitation function of the fuzzy rule with the sequence number of k, wherein the sequence numbers of the fuzzy rule are sequentially arranged according to a reserved sequence, and M represents the number of the most recently input training samples including the currently input training sample with the sequence number of n.
Wherein σ
kRadius of excitation function, mu, of fuzzy rule with index k
kRepresenting the center of the excitation function of the fuzzy rule with index k.
Wherein the content of the first and second substances,
the radius of the excitation function representing the training sample of the current input corresponding to the added fuzzy rule,
the training samples representing the current input correspond to the center of the excitation function of the added fuzzy rule.
Preferably, the setting submodule includes:
and the first setting unit is used for setting a parameter matrix of the fuzzy rule which is correspondingly added to the currently input training sample by adopting a first program group.
Wherein the parameter matrix of the fuzzy rule added corresponding to the currently input training sample is
The first process group is
And the second setting unit is used for setting the center of the excitation function of the added fuzzy rule corresponding to the currently input training sample by adopting a second equation.
Wherein the second equation is
And the third setting unit is used for setting the radius of the excitation function of the fuzzy rule which is added corresponding to the currently input training sample by adopting a third program.
Wherein the third equation is
κ denotes an overlap factor.
Preferably, the system of equations of the least squares recursive error method is:
wherein, a
nA back-piece parameter matrix, y, representing a fuzzy rule with a sequence number n
n-1Indicates the standard type of the foreign body defect corresponding to the training sample with the sequence number of n-1,
indicates the type of the foreign body defect output by the foreign body defect classification model when the training sample with the sequence number of n-1 is input into the foreign body defect classification model, H
0Denotes a preset initial value, H
nRepresenting an input x
nThe output vector of the time-normalized layer,
xne=[1,xn]Tand I represents an identity matrix.
Preferably, the second judgment sub-module includes:
and the second judging unit is used for judging whether the existing fuzzy rule meets the deletion criterion.
And the deleting unit is used for deleting the existing fuzzy rule if the deleting criterion is met.
Wherein the deletion criterion is Emin f(k)<ep,Emin f(k) Representing the influence value of the fuzzy rule with the sequence number k, epIndicating a deletion threshold.
For the device embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, refer to the partial description of the method embodiment.
In summary, the system for detecting the foreign object defect inside the GIS device according to the embodiment of the present invention uses the extended continuous adaptive fuzzy inference system with good classification rate obtained by training as the classification model of the foreign object defect, and inputs the sound signal obtained by the electronic auscultation technique into the classification model of the foreign object defect to obtain the type of the foreign object defect, thereby having higher accuracy and efficiency.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.