CN114997217A - Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM - Google Patents
Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM Download PDFInfo
- Publication number
- CN114997217A CN114997217A CN202210542356.2A CN202210542356A CN114997217A CN 114997217 A CN114997217 A CN 114997217A CN 202210542356 A CN202210542356 A CN 202210542356A CN 114997217 A CN114997217 A CN 114997217A
- Authority
- CN
- China
- Prior art keywords
- wolf
- elm
- igwo
- transformer
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000003745 diagnosis Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000000605 extraction Methods 0.000 claims abstract description 14
- 239000013598 vector Substances 0.000 claims abstract description 11
- 238000002474 experimental method Methods 0.000 claims abstract description 9
- 230000003993 interaction Effects 0.000 claims abstract description 7
- 241000282461 Canis lupus Species 0.000 claims description 102
- 238000005457 optimization Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 17
- 210000002569 neuron Anatomy 0.000 claims description 16
- 230000000739 chaotic effect Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 12
- 238000004891 communication Methods 0.000 claims description 11
- 238000009826 distribution Methods 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 230000006399 behavior Effects 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000013459 approach Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 230000002195 synergetic effect Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 241000282421 Canidae Species 0.000 claims description 2
- 230000009471 action Effects 0.000 claims description 2
- YZCKVEUIGOORGS-IGMARMGPSA-N Protium Chemical compound [1H] YZCKVEUIGOORGS-IGMARMGPSA-N 0.000 claims 1
- 230000000007 visual effect Effects 0.000 claims 1
- 238000004804 winding Methods 0.000 description 11
- 230000008901 benefit Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000012544 monitoring process Methods 0.000 description 6
- 230000001133 acceleration Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 4
- 239000003990 capacitor Substances 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical group [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000002405 diagnostic procedure Methods 0.000 description 2
- 238000004146 energy storage Methods 0.000 description 2
- 239000003381 stabilizer Substances 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000004836 empirical method Methods 0.000 description 1
- 238000004868 gas analysis Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
- 230000000638 stimulation Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
- G06K7/00—Methods or arrangements for sensing record carriers, e.g. for reading patterns
- G06K7/0008—General problems related to the reading of electronic memory record carriers, independent of its reading method, e.g. power transfer
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S40/00—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them
- Y04S40/12—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them characterised by data transport means between the monitoring, controlling or managing units and monitored, controlled or operated electrical equipment
- Y04S40/126—Systems for electrical power generation, transmission, distribution or end-user application management characterised by the use of communication or information technologies, or communication or information technology specific aspects supporting them characterised by data transport means between the monitoring, controlling or managing units and monitored, controlled or operated electrical equipment using wireless data transmission
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a transformer fault diagnosis method based on an RFID sensing tag and IGWO-ELM, which comprises the following steps: s1: building an RFID sensing label transformer signal acquisition experiment platform; s2: the experimental platform carries out information interaction with the RFID reader through radio waves and is responsible for transmitting data returned by the label to an upper computer; s3: the upper computer takes the collected vibration signals as finally classified feature vectors after the vibration signals pass through a sparse depth confidence network SDBN; s4: sending the training set into an improved extreme learning machine IGWO-ELM for training; s5: and testing the trained improved extreme learning machine IGWO-ELM by using the test set, analyzing the diagnosis result and displaying the result in a display interface of the upper computer. The method solves the problems of complex operation, high power consumption and the like of the traditional transformer fault diagnosis method and the problem that the traditional characteristic extraction method is not suitable for extracting the characteristics of the sudden fault signal and the non-stable fault signal in the transformer.
Description
Technical Field
The invention relates to the technical field of transformer fault diagnosis, in particular to a transformer fault diagnosis method based on an RFID sensing tag and IGWO-ELM.
Background
The power transformer is used as a key device of a power supply system, and the running state of the power transformer is directly related to the safety and reliability of the whole power supply system. In long-term operation, mechanical structure faults often occur to a transformer winding and an iron core, and great potential safety hazards exist. Therefore, the method has great significance for monitoring the state and diagnosing the fault of the transformer winding and the iron core.
The existing transformer state online monitoring technology mainly includes Frequency Response Analysis (FRA) and Dissolved Gas Analysis (DGA) in oil. The FRA needs to be electrically connected with a power system, so that the integral anti-interference capability is poor; the DGA does not need to be electrically connected, but cannot determine the specific location and extent of the winding core fault. The transformer fault diagnosis technology based on the vibration signal only needs to attach the vibration sensor to the surface of the wall of the oil tank, does not need to be directly electrically connected with a power grid, has the advantages of online monitoring, high sensitivity, easiness in operation and the like, and gradually becomes a research hotspot at home and abroad in recent years. The traditional vibration signal acquisition mode is that the sensor vibration data is transmitted to the background through a data line, the cost is increased by wired transmission, regular maintenance is needed, and the problems can be effectively avoided by wireless transmission. The existing mainstream wireless technologies mainly include bluetooth (Blue-tooth), Zigbee, Wireless Local Area Network (WLAN) and the like, and these technologies establish a network transmission node based on an auxiliary power supply, so that the cost is high, the power consumption is high, the battery life is also limited, and the technology is not suitable for being applied to a long-term supervision environment. Radio Frequency Identification (RFID) is an advanced data acquisition and automatic Identification technology, a passive RFID sensing tag formed by fusing a sensor and a passive RFID tag adopts a backscattering working mechanism, and the RFID sensing tag has the advantages of simple circuit structure, low cost, low power consumption and the like, and is suitable for long-term state monitoring environments.
Passive RFID tags have limited communication performance due to the radio frequency energy received by the antenna. The single-antenna RFID tag receives radio frequency energy for rectifying circuit energy conversion and data communication, and the radio frequency energy is reduced along with the increase of the working distance, so that the normal communication between the tag and the reader is influenced. Dunfrom et al propose a dual-port tag scheme, where ports are connected to an energy-extracting antenna and a communication antenna, respectively, and this study verifies the feasibility of multi-antenna tag applications.
Methods such as Artificial Neural Network (ANN), Support Vector Machine (SVM), Naive Bayes (NB) and the like are widely applied to data processing in transformer state monitoring, and certain achievements are obtained. An Extreme Learning Machine (ELM) is a novel single hidden layer forward neural network with excellent performance, randomly generates input weights and thresholds, does not need to be adjusted, and has the characteristics of high Learning speed, good generalization performance and the like. However, the correlation studies are mainly based on empirical methods, and the number of the ELM hidden layer neurons can be determined through multiple attempts, so that the theoretical basis of the system is lacked. The literature adopts a Grey wolf algorithm (GWOlfoptimizer, GWOO) to optimize ELM, so that the diagnosis precision of the algorithm is improved to a certain extent, but the convergence speed of GWO is slowed down when the optimal solution is approached, the precision is lowered, and the local optimization is easy to fall into.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art. Therefore, the invention aims to provide a transformer fault diagnosis method based on an RFID sensing tag and IGWO-ELM, which can improve the transformer winding core fault diagnosis precision and reduce the diagnosis time.
According to the transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM, the method comprises the following steps:
s1: building an RFID sensing label transformer signal acquisition experiment platform;
s2: the experimental platform carries out information interaction with the RFID reader through radio waves, after the RFID sensing tag collects a vibration signal of the transformer, the RFID reader is responsible for transmitting a command of the upper computer to the tag, and meanwhile, the RFID reader is responsible for transmitting data returned by the tag to the upper computer; the upper computer performs data preprocessing after acquiring the vibration signals of the transformer, and simulates 8 states generated in the actual use process of the transformer by classifying and sorting the vibration signal data;
s3: the upper computer takes the collected vibration signals as finally classified feature vectors after the vibration signals pass through a sparse depth confidence network (SDBN), and sample data are divided into a training set and a test set according to the proportion of 5: 3;
s4: sending the training set into an improved extreme learning machine IGWO-ELM for training;
s5: and testing the trained improved extreme learning machine IGWO-ELM by using the test set, analyzing the diagnosis result and displaying the result in a display interface of the upper computer.
Preferably, the specific method steps of the experiment platform performing information interaction with the RFID reader through radio waves in step S2 are as follows:
s21: collecting a vibration signal of the transformer by using an RFID sensing tag attached to the outer wall of the transformer;
s22: the energy-taking antenna on the RFID sensing tag receives radio frequency energy generated by radio waves from the RFID reader, the driving circuit works and sends the acquired vibration signal data out by the communication antenna on the RFID sensing tag in a radio wave mode;
s23: the reader receives the data returned from the tag, transmits the data to the upper computer of the system layer through a data line, and the upper computer completes the subsequent data processing process.
Preferably, the specific method steps of using the acquired vibration signal as the finally classified feature vector after passing through the sparse depth confidence network SDBN by the upper computer in step S3 are as follows:
s31: respectively passing the acquired transformer vibration signals through a sparse depth confidence network SDBN to extract deep fault characteristics, wherein the sparse depth confidence network SDBN comprises the following specific construction steps:
sparsity constraint of restricted boltzmann machine RBM:
a j (x i )＝f(w j ·x i +b j )
in the formula: a is j (x i ) For the activation function of hidden layer neuron j,
mean activation value for hidden layer neurons; in order to meet the sparsity constraint condition, the mean activation value of the neurons in the hidden layer is required to be close to 0, so that a sparsity parameter rho is introduced, the target value is close to 0, and KL divergence formula is used for measuring
From p to p, such that
M is the number of neurons in the hidden layer, f is the activation function, i represents the number of signal sample sets, w represents the weight matrix of the network, b represents the bias parameter, x represents the signal samples, n represents the parameter of the summation formula, a j (x i ) Representing the activation function:
by introducing a sparse mechanism into the RBM, constructing an SRBM sparse depth belief network model to enable the RBM to learn sparse expression, and constructing an SDBN (software development bus) on the basis, wherein the structure is formed by connecting a plurality of SRBMs in series, the hidden layer of the last SRBM is the visible layer of the next SRBM, and the output of the last SRBM is the input of the next SRBM;
s32: and by the characteristic extraction operation of the SDBN model, the extracted characteristic vector is used as input data of a subsequent classifier.
Preferably, the specific method steps for improving the extreme learning machine IGWO-ELM in step S4 are as follows:
s41: GWO constructing a gray wolf algorithm model;
s42: initializing the population by utilizing a chaotic factor optimization algorithm;
s43: introducing a normal cloud model as a wolf group updating mechanism in the gray wolf attack stage;
s44: constructing an ELM extreme learning machine algorithm model;
s45: and (3) improving the grey wolf algorithm optimization extreme learning machine, and establishing an IGWO-ELM model.
Preferably, the specific method steps for constructing GWO gray wolf algorithm model in step S41 are as follows:
s411: the method comprises the following steps of (1) social level layering, wherein a social level layering model of the gray wolf is constructed in a gray wolf algorithm, the fitness of individuals in a wolf group needs to be calculated, the three wolfs with the best fitness are respectively marked as alpha, beta and delta, and the rest wolfs are marked as wolfs;
s412: when the gray wolf finds the prey, the distance D between the gray wolf and the prey is calculated, then the gray wolf continuously updates the position and gradually approaches to the prey, and the mathematical formula of the behavior is expressed as follows:
D＝C·X p (t)-X(t)
X(t+1)＝X p (t)-A·D
A＝2a·r1-a
C＝2·r2
in the formula: x (t) is the current location of the wolf; x p (t) is the present position of the prey; t is the number of iterations; a and C are synergistic coefficients; r is 1 And r 2 Is a random number, range [0, 1]](ii) a a is a convergence factor;
s413: hunting attack, after the gray wolf finds the prey, the beta wolf and the delta wolf surround the prey under the instruction of the decision maker alpha wolf; since the precise position of the prey cannot be determined, assuming that α, β, δ wolf has strong recognition capability for the target, in each recognition action, the positions of other individuals are updated according to the position information of α, β, δ wolf, and the specific distance formula is as follows:
D α ＝|C 1 ·X α -X|
D β ＝|C 2 ·X β -X|
X 1 ＝|X α -A 1 ·D α |
X 2 ＝|X β -A 2 ·D β |
X 3 ＝|X δ -A 3 ·D δ |
in the formula: d α, D β, D δ are distances between α, β, δ wolf and other individuals, respectively; x α ，X β ，X δ The positions of the alpha, beta, delta wolves, respectively; c 1 ，C 2 ，C 3 Is a random number; x is the current position; x (t+1) Is the final position of the wolf, X 1 、X 2 、X 3 Is the current latest position of the wolf of alpha, beta, delta, A 1 、A 2 、A 3 、C 1 、C 2 、C 3 Is a coefficient of synergy;
s414: when A is more than 1, the Grey wolf carries out global search, the search coefficient C can also be used for searching, random weight is mainly provided for the target, random search behavior is facilitated, and C takes a value between [0 and 2 ].
Preferably, in step S42, the chaos factor optimization algorithm is used to initialize the population according to the following specific formula:
wherein:
is a chaotic sequence; mu e (0, 2)]Is a chaotic parameter; i is the population number and j is the chaotic variable serial number.
Preferably, the specific method steps of introducing the normal cloud model as the wolf pack update mechanism into the grayish wolf attack stage in the step S43 are as follows:
assuming that a set Q is { c }, Q is a domain with a constant value c, a random number mu with a stable tendency exists in any element c in a fuzzy set A in the domain Q A (c) Is called c to mu A (c) The cloud model fitting formula is as follows:
ζ＝max p(x)
wherein: ex is an expected value parameter; en is an entropy parameter; he is a super entropy parameter; is the fitting correction; p (x) is the peak probability of the cloud drop x probability score, f (x) is the cloud model fitting formula, ζ is the fitting correction amount, E' n Is a normal random number, p (x) is the peak probability of the cloud drop x probability score;
the distribution of random variables formed by all cloud droplets x on a domain of discourse Q is called a normal cloud model, the normal cloud model is characterized by expected Ex, entropy En and super entropy He, the expected Ex can represent a qualitative concept and is an expectation of prediction errors in the domain of discourse distribution; the entropy En represents the uncertainty of the prediction error, i.e. the amplitude fluctuation range of the error; the super-entropy He is an uncertain measure of the entropy En and represents the concentration degree of prediction errors, and the calculation formulas are respectively as follows:
in the formula: i is a sample serial number; n is a sample total book; e.g. of the type i For prediction error, S 2 Is the sum of squares between the predicted value and the expected value;
the normal cloud generator is an algorithm obeying a normal cloud model, and generates one cloud droplet every time the generator operates until an expected cloud droplet number Nd is generated, wherein the formula of the normal cloud generator is as follows:
X[x 1 ,x 2 ,…,x Nd ]＝Gnc(Ex,En,He,Nd)
after the Tent mapping is introduced in the population initialization, taking a normal cloud model as an updating mechanism of the wolf pack position, taking the position of the current optimal individual as an expected Ex, and taking a position updating formula as follows, wherein position _ best is the position of the current optimal individual;
position'＝Gnc(position_best,En,He,Nd)
updating the distance range of the optimal individual position by adjusting En, adjusting He to control the dispersion degree of the wolf colony position, and gradually approaching the position range of the wolf colony to a prey according to the prey process of tracking, surrounding and attacking the wolf colony, so that the values of En and He can be adaptively adjusted as follows:
in the formula, En represents entropy, He represents super-entropy, omega epsilon (0, 1) is a random number, t is the current iteration number, tau and xi are positive integers, and max iter is the maximum iteration number.
Preferably, the specific method for constructing the ELM algorithm model in step S44 includes the following steps:
assuming that the training samples have N arbitrary samples (xi, yi), for a single hidden layer feedforward neural network with N hidden layer nodes can be expressed as:
the corresponding matrix form is:
Hβ＝Y
the loss function is expressed as:
preferably, in step S45, the grey wolf algorithm is improved to optimize the extreme learning machine, and the specific method for establishing the IGWO-ELM model includes the following steps:
introducing IGWO into ELM parameter optimization process, and optimizing ELM network by using dynamic optimization characteristics; initializing a gray wolf algorithm parameter, and training by randomly giving an initial network weight and a threshold by the ELM; updating the spatial position of the wolf with error feedback, and retraining the network weight and the threshold value by the ELM; and (4) taking the object with the lowest misjudgment rate as a prey to capture, and continuously optimizing the fitness value until a global optimal solution meeting the conditions, namely the optimal weight and the threshold of the ELM model, is obtained to form an optimal recognition model.
The beneficial effects of the invention are as follows:
(1) the invention provides a double-antenna RFID sensing tag which can stably collect and transmit vibration signals of a transformer and has the characteristics of low cost, low power consumption, good real-time performance and the like;
(2) compared with the traditional signal-based feature extraction method, the fault signal feature extraction method of the SDBN has obvious advantages in mixed fault state classification, can realize high aggregation of the same state, obviously separates different states and has strong discrimination;
(3) the ELM has good nonlinear fitting capability, the input weight Wi and the bias bi are randomly generated without adjustment, and the condition that the random number is 0 is generated by the traditional ELM initialization generation weight and bias, so that part of hidden layers are invalid, the generalization performance is reduced, and the local optimization is easy to fall into. Fourthly, the invention can enable a transformer manager to clearly master the operation information and the historical fault information in real time by using the interfaces such as the real-time state information of the inverter, the historical fault record display and the like created by the C.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
fig. 1 is a flowchart of a transformer fault diagnosis method based on an RFID sensor tag and IGWO-ELM according to the present invention;
FIG. 2 is a schematic diagram of a circuit according to the present invention;
FIG. 3 is a flow chart of a DBN proposed by the present invention;
FIG. 4 is a flow chart of ELM according to the present invention;
FIG. 5 is a structural diagram of IGWO-ELM according to the present invention;
FIG. 6 is a graph showing the relationship between the number of nodes in hidden layers of different networks and the error according to the present invention;
FIG. 7 is a comparison of the effect of the SDBN extraction method proposed by the present invention;
FIG. 8 is a comparison of the effects of the conventional DBN extraction method proposed by the present invention;
FIG. 9 is a comparison graph of the effect of the wavelet transform extraction method proposed by the present invention;
FIG. 10 is a bar graph of a test set training phase proposed by the present invention;
FIG. 11 is a bar graph of a test set testing phase proposed by the present invention;
fig. 12 is a screenshot of a real-time monitoring interface of a transformer signal diagnosis system according to the present invention.
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 only a part of the embodiments of the present invention, and not all of the embodiments.
Referring to fig. 1, a transformer fault diagnosis method based on an RFID sensing tag and an IGWO-ELM includes the following steps:
(1) building an RFID sensing label transformer signal acquisition experiment platform;
(2) the experimental platform carries out information interaction with the RFID reader through radio waves, and after vibration signals are collected, the RFID reader is responsible for transmitting commands of the upper computer to the tags and transmitting data returned by the tags to the upper computer; carrying out data preprocessing after collecting vibration signals of the transformer, and simulating 8 states generated in the actual use process of the transformer by classifying and sorting the vibration signal data;
(3) the upper computer takes the collected vibration signals as finally classified feature vectors after the vibration signals pass through a sparse depth confidence network (SDBN), and sample data are divided into a training set and a test set according to the proportion of 5: 3;
(4) sending the training set into an improved extreme learning machine IGWO-ELM for training;
(5) and testing the trained improved extreme learning machine IGWO-ELM by using the test set, analyzing the diagnosis result and displaying the result in a display interface of the upper computer.
The experiment platform carries out information interaction through between radio wave and the RFID reader in step (2), and after gathering vibration signal, the RFID reader is responsible for sending the order of host computer to the label, is responsible for simultaneously sending the data that the label returned to the host computer, specifically indicates:
(2a) collecting a vibration signal of the transformer by using an RFID sensing tag attached to the outer wall of the transformer;
(2b) an energy-taking antenna on the RFID sensing tag receives radio frequency energy generated by radio waves from a reader, a driving circuit works, and collected vibration signal data are sent out by a communication antenna on the sensing tag in a radio wave mode;
(2c) the reader receives the data returned from the tag, transmits the data to the upper computer of the system layer through a data line, and the upper computer completes the subsequent data processing process.
In the step (3), the upper computer takes the collected vibration signals as finally classified feature vectors after the vibration signals pass through a Sparse Deep Belief Network (SDBN), and the method specifically comprises the following steps:
(3a) respectively passing the acquired vibration signals of the fault state of the transformer through a Sparse Deep Belief Network (SDBN), and extracting deep fault characteristics;
(3b) the specific construction steps of the sparse deep belief network SDBN are as follows:
sparsity constraint of restricted boltzmann machine RBM:
a j (x i )＝f(w j ·x i +b j )
in the formula: a is j (x i ) As a function of the degree of activation of hidden layer neurons j,
mean activation value for hidden layer neurons; in order to satisfy the sparsity constraint condition, the mean activation value of the neurons in the hidden layer is required to be close to 0, so that the sparsity parameter p is introduced, and the target value is close to 0 generally. KL (Kullback-Leibler divergence) is mostly applied to probability theory or information theory, which can be called relative entropy (relative entropy), and is measured by using KL divergence formula
From p to p, such that
m is the number of neurons in the hidden layer:
a Sparse mechanism is introduced into the RBM to construct an SRBM (Sparse-RBM) model, so that the RBM can learn Sparse expression, and the SDBN is constructed on the basis. The structure is formed by connecting a plurality of SRBMs in series, the hidden layer of the last SRBM is the visible layer of the next SRBM, the output of the last SRBM is the input of the next SRBM, and the characteristic representation of increasing sparse constraint can effectively improve the target recognition performance of the DBN network.
(3c) And using the extracted feature vector as input data of a subsequent classifier through the feature extraction operation of the SDBN model.
The step (4) comprises the steps of improving a grey wolf algorithm to optimize an extreme learning machine and using the improved extreme learning machine to classify faults; the improved wolf algorithm comprises steps (4a) to (4d), and the optimized extreme learning machine specifically comprises step (4 e):
(4a) construction GWO:
(4aa) social ranking layering: the social level hierarchical model of the gray wolf is constructed in the gray wolf algorithm, the fitness of individuals in wolf groups needs to be calculated, the three wolfs with the best fitness are respectively marked as alpha, beta and delta, and the rest wolfs are marked as omega wolfs.
(4ab) surrounding prey: when the gray wolf finds the prey, firstly, the distance D between the gray wolf and the prey is calculated, then the gray wolf continuously updates the position and gradually approaches to the surrounding prey, and the mathematical formula of the behavior is expressed as follows:
D＝C·X p (t)-X(t)
X(t+1)＝X p (t)-A·D
A＝2a·r 1 -a
C＝2·r 2
in the formula: x (t) is the current location of the wolf; x p (t) is the current position of the prey; t is the number of iterations; a and C are synergistic coefficients; r is 1 And r 2 Is a random number, range [0, 1]](ii) a a is the convergence factor.
(4ac) hunting attacks: after the gray wolf finds the prey, the beta wolf and the delta wolf surround the prey under the instruction of the decision maker alpha wolf; since the precise location of the prey cannot be determined, it is assumed that α, β, δ wolfs have strong recognition ability for the target. In each recognition behavior, the positions of other individuals are updated according to the position information of alpha, beta and delta wolfs, and the specific distance formula is as follows:
D α ＝|C 1 ·X α -X|
D β ＝|C 2 ·X β -X|
X 1 ＝|X α -A 1 ·D α |
X 2 ＝|X β -A 2 ·D β |
X 3 ＝|X δ -A 3 ·D δ |
in the formula: d α, D β, D δ are distances between α, β, δ wolf and other individuals, respectively; x α ，X β ，X δ The positions of alpha, beta, delta wolf, respectively; c 1 ，C 2 ，C 3 Is a random number; x is the current position; x (t+1) Is the final position of the w wolf.
(4ad) search for prey: when A is larger than 1, the Grey wolf carries out global search, the search coefficient C can also be used for searching, random weight is mainly provided for the target, random search behavior is facilitated, and C takes a value between [0 and 2 ].
(4b) Initializing a chaotic optimization algorithm population: the chaotic sequence has good randomness, ergodicity and regularity, and the basic principle is that the chaotic sequence is generated between values [0, 1] through a mapping relation and is converted into an individual search space. The more uniform the distribution of the initial population in the space, the more beneficial the optimization process, compared with other mappings, Tent mapping has more balanced sequence, and the formula is as follows:
wherein:
is a chaotic sequence; mu e (0, 2)]Is a chaotic parameter; i is the population number and j is the chaotic variable serial number.
(4c) The grey wolf attack stage introduces a normal cloud model as a wolf group updating mechanism: the cloud model is an uncertain conversion tool for realizing qualitative concepts and quantitative values, can well describe data randomness and fuzziness, and is most consistent with natural random probability distribution.
Let the existence of the set Q ═ c } and Q be the domain of constant value c. There is a stable trend random number mu for any element c in the fuzzy set A in the discourse domain Q A (c) Is called c to mu A (c) Degree of membership. The cloud model fitting formula is as follows:
E′ n ～N(E n ,H e )
ζ＝max p(x)
wherein: ex is an expected value parameter; en is an entropy parameter; he is a super entropy parameter; is the fitting correction; p (x) is the peak probability of the cloud drop x probability score.
The distribution of random variables made up of all cloud droplets x over the domain of discourse Q is referred to as the normal cloud model. The normal cloud model is characterized by expected Ex, entropy En and super-entropy He, wherein the expected Ex can represent qualitative concepts most and is an expectation of prediction errors in domain distribution; the entropy En represents the uncertainty of the prediction error, i.e. the amplitude fluctuation range of the error; the super-entropy He is an uncertain measure of the entropy En, representing the degree of concentration of the prediction error. The calculation formulas are respectively as follows:
in the formula: i is a sample serial number; n is a sample total book; e.g. of a cylinder i Is the prediction error.
The normal cloud generator is an algorithm obeying a normal cloud model, and generates one cloud drop every time the generator operates until an expected cloud drop number Nd is generated, wherein the formula of the normal cloud generator is as follows:
X[x 1 ,x 2 ,…,x Nd ]＝Gnc(Ex,En,He,Nd)
after the Tent mapping is introduced into the population initialization, the normal cloud model is used as an updating mechanism of the wolf pack position, the position of the current optimal individual is taken as expected Ex, the position updating formula is as follows, and position _ best is the position of the current optimal individual.
position'＝Gnc(position_best,En,He,Nd)
And updating the distance range of the optimal individual position by adjusting En, and adjusting He to control the dispersion degree of the wolf group position. According to the predation process of tracking, surrounding and attacking the wolf pack, the position range of the wolf pack from a prey gradually approaches, so that the values of En and He can be adaptively adjusted as follows:
(4d) constructing an ELM: ELM is a new type of fast learning algorithm, which consists of input layer, hidden layer and output layer, the input layer and hidden layer, and the neurons between hidden layer and output layer are connected by full connection layer. For the single hidden layer neural network, the ELM can randomly initialize the input weight and the bias and obtain the corresponding output weight, thereby not only reducing the training time, but also obtaining the global optimal solution.
Assuming that the training samples have N arbitrary samples (xi, yi), for a single hidden layer feedforward neural network with N hidden layer nodes can be expressed as:
the corresponding matrix form is:
Hβ＝Y
the loss function is expressed as:
for model solving, it is desirable to find a specific set of solutions such that the loss function is minimized; different from the traditional function approximation method, the output matrix H can be kept unchanged only by setting appropriate hidden layer nodes in the training process in the extreme learning machine network structure, so that the training of the single hidden layer neural network is converted into a least square solution for solving a linear system H beta-Y.
(4e) Improving a grey wolf algorithm optimization extreme learning machine, and establishing an IGWO-ELM model: in order to solve the problems that the random number is 0 due to the fact that weight and bias generated by EL M initialization generate, part of hidden layers are invalid, and generalization performance is reduced, IGWO is introduced into the parameter optimization process of ELM, and an ELM network is optimized by using the characteristic of dynamic optimization.
Initializing a gray wolf algorithm parameter, and training by randomly giving an initial network weight and a threshold by the ELM; updating the spatial position of the wolf with error feedback, and retraining the network weight and the threshold value by the ELM; and (4) taking the object with the lowest misjudgment rate as a prey to capture, and continuously optimizing the fitness value until a global optimal solution meeting the conditions, namely the optimal weight and the threshold of the ELM model, is obtained to form an optimal recognition model.
As shown in fig. 2, the present system includes:
the RFID label chip adopts a MonzaX-8K chip and is supported by I 2 C, performing read-write operation on the memory;
the communication antenna adopts a microstrip antenna, so that the communication capability is improved, and the anti-interference capability is strong;
the matching network adopts a high-quality-factor ceramic dielectric trimming capacitor and an adjustable radio frequency inductor to complete impedance matching and power matching between the energy-taking antenna and the circuit, so that the received signal power of the antenna is maximized.
The rectification circuit adopts a four-stage boosting rectification circuit and converts the radio frequency energy acquired by the energy acquisition antenna into direct current voltage. The energy storage capacitor is formed by connecting two 220 mu F capacitors in parallel, so that the power requirements of continuous work of the sensing circuit and intermittent work of the sensing tag are met.
The voltage stabilizing circuit adopts the TPS780180300DRVR low drop-out voltage stabilizer to obtain stable direct current voltage output, and the unstable influence of voltage caused by radio frequency energy fluctuation is reduced.
The vibration signals of the transformer are collected through an ADXL372 three-axis acceleration sensor.
The MCU adopts MSP430FR5992 to control the working process of the RFID sensing tag, and the working process can be controlled through I 2 And C, reading the acquired transformer vibration signal data from the acceleration sensor by using the bus.
And the upper computer is used for processing the acquired data, and conveying the characteristic vectors to the improved ELM for classification to obtain a diagnosis result.
The RFID sensing label module is composed of a radio frequency module, an energy management module and a digital module. The radio frequency module is composed of a communication antenna and an RFID chip and is used for completing the demodulation and modulation functions of communication between the tag and the reader. The energy management module is composed of an energy-taking antenna, a matching network, a rectifying circuit, an energy storage capacitor and a voltage stabilizer and is responsible for providing power supply voltage for normal work of the sensing tag. The digital module is composed of a microcontroller, I 2 The C bus and the three-axis acceleration sensor are used for controlling the whole working process of the RFID sensing label.
Through a designed experiment platform, 8 different transformer vibration signals are collected, namely 6 single states and 2 mixed fault states, so that the invention can be researched aiming at the following 8 different transformer fault states. Transformer state classification is shown in table 1:
TABLE 1 Fault status Classification
Through attaching RFID sensing label in the transformer outer wall, adopt triaxial acceleration sensor to gather the vibration signal data of transformer to regard the first 128 Fourier coefficients of vibration signal as raw data, this article transformer fault state can be divided into 8 kinds in table 2. And acquiring 240 groups of data in each state, randomly selecting 150 groups of data as training data and 90 groups of data as test data, and establishing a transformer characteristic extraction model through the constructed SDBN network.
To study the relationship between the number of layers of the network structure and the error level, after initializing the SDBN network, it can be seen from fig. 6 through a plurality of experiments: when the number of layers reaches 4, the error is the lowest, the number of layers continues to increase, and the error gradually increases, which indicates that the SDBN network may generate an overfitting phenomenon, so that the SDBN adopts a 4-layer hidden layer structure. Optimizing the number of 4-layer hidden layer neurons of the SDBN network by an IGWO algorithm, setting the maximum hidden layer node number to be 500 and the minimum hidden layer node number to be 100, and obtaining an optimized node structure of 426-326-275-401. As the input data is 128 dimensions and the corresponding transformer state categories are 8, the SDBN network structure is set to 128-. In addition, selecting sigmoid as a stimulation function of the neuron; the maximum iteration number of each layer of RBM is 110; the learning rate of the weight is set to 0.01, and the network parameters of the SDBN are shown in table 2:
table 2 network parameters of SDBN
Fig. 7, 8 and 9 are KPCA visualization diagrams of features extracted by three methods, namely SDBN, traditional DBN and wavelet transformation, respectively. It can be seen from the figure that the characteristics extracted by the traditional DBN method overlap the winding deformation, winding nesting and core deformation faults in a single fault state, and are obviously inferior to the SDBN method of the present invention in the classification of the mixed fault states 7 and 8, and it can be seen that the distribution range of various state characteristics is large and the aggregation degree is low; the characteristics extracted by the wavelet analysis method are obviously overlapped in the normal state and the winding deformation fault in the single fault and the mixed fault states 7 and 8, and the characteristics in different states do not have strong separability, so that the characteristics are not beneficial to further identification of a subsequent diagnosis method.
In conclusion, the features extracted by the SDBN method have obvious advantages in mixed fault state classification, the same states are highly aggregated, different states have obvious separation effect, and the discrimination is strong. The improved sparse deep confidence network has better effect than the traditional signal-based feature extraction method when being applied to vibration signal data extraction with large quantity, high dimensionality, complex components and low signal-to-noise ratio, and is beneficial to improving the accuracy of subsequent fault diagnosis.
After the SDBN method is applied to feature extraction of training data of the transformer vibration signal, fault feature data are obtained according to the following steps of 5:3, dividing the ratio into a training set and a test set, sending the training set and the test set to an IGWO-ELM fault diagnosis model for fault state identification, and obtaining diagnosis results of various fault states in a table 2.
TABLE 2 diagnosis results of various types of faults
Figures 10, 11 are test set test results showing that 1 winding nesting sample was erroneously identified as a winding loosening fault; and the accuracy of the winding deformation in the mixed fault state and the iron core loosening fault state is 98.9 percent, and 1 sample is identified as a single winding deformation fault. The overall correct recognition rate for this diagnosis was 99.73%.
Table 3 shows the comparison of the performance of the five diagnostic methods, and it can be seen from the table that, in the five fault diagnosis models, the average fault diagnosis rate of only the training data and the test data of the IGWO-ELM model herein reaches 100%. The average fault diagnosis rate of the IGWO-ELM and the IGWO-SVM is also obviously higher than the model fault diagnosis rate of the ELM and the SVM, mainly because the ELM and the SVM models are artificially set according to experience summary in network parameter selection and have higher randomness; compared with an GWO-ELM diagnostic model, the IGWO-ELM diagnostic model has little difference in diagnostic time of test data, but the average fault diagnosis rate of the IGWO-ELM diagnostic model in the test data is superior to that of GWO-ELM, which shows that the wolf optimization algorithm based on the adaptive normal cloud model provided by the invention achieves the effect of improving the diagnostic accuracy on the selection of the optimal parameters of the diagnostic model. In conclusion, the method provided by the invention has high diagnosis accuracy and short diagnosis time, and is suitable for on-line fault diagnosis of the transformer.
TABLE 3 comparison of the Performance of the five diagnostic methods
In summary, the invention provides a transformer fault diagnosis system based on an RFID sensing tag and an IGWO-ELM, which collects vibration signal data of a transformer through a designed dual-antenna RFID sensing tag, performs deep feature extraction on the original vibration signal data of the transformer by using a sparse deep belief network, and performs fault diagnosis on the state of the transformer by using an extreme learning machine model optimized by a self-adaptive normal cloud-based gray wolf algorithm. The system shows that the designed RFID sensing tag can stably collect and transmit the vibration signal of the transformer, and has the advantages of low cost, low power consumption and the like; compared with the traditional signal-based feature extraction method, the SDBN has obvious advantages in mixed fault state classification, can realize high aggregation of the same state, obviously separates different states and has strong discrimination; the transformer fault diagnosis method of IGWO-ELM integrally achieves high diagnosis precision and realizes efficient and accurate state recognition.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (9)
1. A transformer fault diagnosis method based on RFID sensing labels and IGWO-ELM is characterized by comprising the following steps:
s1: building an RFID sensing label transformer signal acquisition experiment platform;
s2: the experimental platform carries out information interaction with the RFID reader through radio waves, after the RFID sensing tag collects a vibration signal of the transformer, the RFID reader is responsible for transmitting a command of the upper computer to the tag, and meanwhile, the RFID reader is responsible for transmitting data returned by the tag to the upper computer; the upper computer performs data preprocessing after acquiring the vibration signals of the transformer, and simulates 8 states generated in the actual use process of the transformer by classifying and sorting the vibration signal data;
s3: the upper computer takes the collected vibration signals as finally classified feature vectors after the vibration signals pass through a sparse depth confidence network (SDBN), and sample data are divided into a training set and a test set according to the proportion of 5: 3;
s4: sending the training set into an improved extreme learning machine IGWO-ELM for training;
s5: and testing the trained improved extreme learning machine IGWO-ELM by using the test set, analyzing the diagnosis result and displaying the result in a display interface of the upper computer.
2. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 1, wherein the specific method steps of the experiment platform performing information interaction with the RFID reader through radio waves in step S2 are as follows:
s21: collecting a vibration signal of the transformer by using an RFID sensing tag attached to the outer wall of the transformer;
s22: the energy-taking antenna on the RFID sensing tag receives radio frequency energy generated by radio waves from an RFID reader, the driving circuit works, and the acquired vibration signal data is transmitted by the communication antenna on the RFID sensing tag in a radio wave mode;
s23: the reader receives the data returned from the tag, and transmits the data to the upper computer of the system layer through the data line, and the upper computer completes the subsequent data processing process.
3. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 1, wherein the specific method steps of the collected vibration signal, which is used as the feature vector of the final classification after passing through the sparse deep belief network SDBN by the upper computer in the step S3 are as follows:
s31: the method comprises the following steps of respectively passing acquired transformer vibration signals through a sparse deep belief network SDBN to extract deep fault characteristics, wherein the sparse deep belief network SDBN comprises the following specific construction steps:
sparsity constraint of restricted boltzmann machine RBM:
a j (x i )＝f(w j ·x i +b j )
in the formula: a is j (x i ) As a function of the degree of activation of hidden layer neurons j,
mean activation value for hidden layer neurons; in order to meet the sparsity constraint condition, the mean activation value of the neurons in the hidden layer is required to be close to 0, so that a sparsity parameter rho is introduced, the target value is close to 0, and KL divergence formula is used for measuring
From p to p, such that
m is the number of neurons in the hidden layer, f is the activation function, i represents the number of signal sample sets, w represents the weight matrix of the network, b represents the bias parameter, x represents the signal samples, n represents the parameter of the summation formula, a j (x i ) Representing the activation function:
constructing an SRBM sparse deep belief network model by introducing a sparse mechanism into an RBM, so that the RBM can learn sparse expression, and constructing an SDBN (software development bridge) on the basis, wherein the structure is formed by connecting a plurality of SRBMs in series, the previous SRBM hidden layer is the next SRBM visual layer, and the previous SRBM output is the next SRBM input;
s32: and by the characteristic extraction operation of the SDBN model, the extracted characteristic vector is used as input data of a subsequent classifier.
4. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 1, wherein the specific method steps for improving the IGWO-ELM in the step S4 are as follows:
s41: GWO constructing a gray wolf algorithm model;
s42: initializing the population by utilizing a chaotic factor optimization algorithm;
s43: introducing a normal cloud model as a wolf group updating mechanism in the gray wolf attack stage;
s44: constructing an ELM extreme learning machine algorithm model;
s45: and (3) improving the grey wolf algorithm optimization extreme learning machine, and establishing an IGWO-ELM model.
5. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 4, wherein the specific method steps for constructing GWO gray wolf algorithm model in step S41 are as follows:
s411: the method comprises the following steps of (1) social level layering, wherein a social level layering model of the gray wolf is constructed in a gray wolf algorithm, the fitness of individuals in a wolf group needs to be calculated, the three wolfs with the best fitness are respectively marked as alpha, beta and delta, and the rest wolfs are marked as wolfs;
s412: when the gray wolf finds the prey, the distance D between the gray wolf and the prey is calculated, then the gray wolf continuously updates the position and gradually approaches to the prey, and the mathematical formula of the behavior is expressed as follows:
D＝C·X p (t)-X(t)
X(t+1)＝X p (t)-A·D
A＝2a·r 1 -a
C＝2·r 2
in the formula: x (t) is the current location of the wolf; x p (t) is the present position of the prey; t is the number of iterations; a and C are synergistic coefficients; r is a radical of hydrogen 1 And r 2 Is a random number, range [0, 1]](ii) a a is a convergence factor;
s413: hunting attack, after the gray wolf finds the prey, the beta wolf and the delta wolf surround the prey under the instruction of the decision maker alpha wolf; since the precise position of the prey cannot be determined, assuming that α, β, δ wolf has strong recognition capability for the target, in each recognition action, the positions of other individuals are updated according to the position information of α, β, δ wolf, and the specific distance formula is as follows:
D α ＝|C 1 ·X α -X|
D β ＝|C 2 ·X β -X|
X 1 ＝|X α -A 1 ·D α |
X 2 ＝|X β -A 2 ·D β |
X 3 ＝|X δ -A 3 ·D δ |
in the formula: d α, D β, D δ are distances between α, β, δ wolf and other individuals, respectively; x α ，X β ，X δ The positions of the alpha, beta, delta wolves, respectively; c 1 ，C 2 ，C 3 Is a random number; x is the current position; x (t+1) Is the final position of the wolf, X 1 、X 2 、X 3 Is the current latest position of alpha, beta, delta wolf, A 1 、A 2 、A 3 、C 1 、C 2 、C 3 Is a synergistic coefficient;
s414: when A is more than 1, the Grey wolf carries out global search, the search coefficient C can also be used for searching, random weight is mainly provided for the target, random search behavior is facilitated, and C takes a value between [0 and 2 ].
6. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 5, wherein the chaotic factor optimization algorithm is used in step S42 to make the specific formula of population initialization as follows:
wherein:
is a chaotic sequence; mu e (0, 2)]Is a chaotic parameter; i is the population number and j is the chaotic variable serial number.
7. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 6, wherein the specific method steps of introducing the normal cloud model as a wolf pack update mechanism in the grey wolf attack stage in the step S43 are as follows:
assuming that a set Q is { c }, Q is a domain with a constant value c, a random number mu with a stable tendency exists in any element c in a fuzzy set A in the domain Q A (c) Is called c to mu A (c) Degree of membership ofThe cloud model fitting formula is as follows:
E′ n ～N(E n ,H e )
ζ＝max p(x)
wherein: ex is an expected value parameter; en is an entropy parameter; he is a super entropy parameter; is the fitting correction; p (x) is the peak probability of the cloud drop x probability score, f (x) is the cloud model fitting formula, ζ is the fitting correction amount, E' n Is a normal random number, p (x) is the peak probability of the cloud drop x probability score;
the distribution of random variables formed by all cloud droplets x on a domain of discourse Q is called a normal cloud model, the normal cloud model is characterized by expected Ex, entropy En and super entropy He, the expected Ex can represent a qualitative concept and is an expectation of prediction errors in the domain of discourse distribution; the entropy En represents the uncertainty of the prediction error, i.e. the amplitude fluctuation range of the error; the super-entropy He is an uncertain measure of the entropy En and represents the concentration degree of prediction errors, and the calculation formulas are respectively as follows:
in the formula: i is a sample serial number; n is a sample total book; e.g. of the type i In order to predict the error, the prediction error is calculated,S 2 is the sum of squares between the predicted value and the expected value;
the normal cloud generator is an algorithm obeying a normal cloud model, and generates one cloud drop every time the generator operates until an expected cloud drop number Nd is generated, wherein the formula of the normal cloud generator is as follows:
X[x 1 ,x 2 ,…,x Nd ]＝Gnc(Ex,En,He,Nd)
after the Tent mapping is introduced in the population initialization, taking a normal cloud model as an updating mechanism of the wolf pack position, taking the position of the current optimal individual as an expected Ex, and taking a position updating formula as follows, wherein position _ best is the position of the current optimal individual;
position'＝Gnc(position_best,En,He,Nd)
updating the distance range of the optimal individual position by adjusting En, adjusting He to control the dispersion degree of the wolf colony position, and gradually approaching the position range of the wolf colony to a prey according to the prey process of tracking, surrounding and attacking the wolf colony, so that the values of En and He can be adaptively adjusted as follows:
in the formula, En represents entropy, He represents super-entropy, omega epsilon (0, 1) is a random number, t is the current iteration number, tau and xi are positive integers, and max iter is the maximum iteration number.
8. The transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 7, wherein the specific method steps for constructing the ELM algorithm model in step S44 are as follows:
assuming that the training samples have N arbitrary samples (xi, yi), for a single hidden layer feedforward neural network with N hidden layer nodes can be expressed as:
the corresponding matrix form is:
Hβ＝Y
the loss function is expressed as:
9. the transformer fault diagnosis method based on the RFID sensing tag and the IGWO-ELM as claimed in claim 8, wherein the specific method steps for establishing the IGWO-ELM model by improving the gray wolf algorithm to optimize the extreme learning machine in step S45 are as follows:
introducing IGWO into the parameter optimization process of ELM, and optimizing the ELM network by using the characteristic of dynamic optimization; initializing a gray wolf algorithm parameter, and training by randomly giving an initial network weight and a threshold by the ELM; updating the spatial position of the wolf with error feedback, and retraining the network weight and the threshold value by the ELM; and (4) taking the object with the lowest misjudgment rate as a prey to perform enclosure catching, and continuously optimizing the fitness value until a global optimal solution meeting the conditions, namely the optimal weight and the threshold of the ELM model, is obtained to form an optimal recognition model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210542356.2A CN114997217B (en) | 2022-05-18 | 2022-05-18 | Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210542356.2A CN114997217B (en) | 2022-05-18 | 2022-05-18 | Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114997217A true CN114997217A (en) | 2022-09-02 |
| CN114997217B CN114997217B (en) | 2024-09-13 |
Family
ID=83027885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210542356.2A Active CN114997217B (en) | 2022-05-18 | 2022-05-18 | Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114997217B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256592A (en) * | 2022-11-28 | 2023-06-13 | 国网山东省电力公司德州供电公司 | Medium-voltage distribution cable latent fault detection method and system |
| CN117150374A (en) * | 2023-10-27 | 2023-12-01 | 国网江西省电力有限公司电力科学研究院 | Transformer fault identification method and system based on feature importance sorting |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101762534B1 (en) * | 2016-02-23 | 2017-07-28 | 호남대학교 산학협력단 | System for remote diagnosing failure and monitoring of electric equipment using low power wide area network |
| CN110132596A (en) * | 2019-04-24 | 2019-08-16 | 昆明理工大学 | A method of the rolling bearing fault diagnosis based on wavelet packet and GWO-SVM |
| CN113125992A (en) * | 2021-04-23 | 2021-07-16 | 合肥工业大学 | NPC three-level inverter fault diagnosis method and system based on DBN |
| CN113923104A (en) * | 2021-12-07 | 2022-01-11 | 南京信息工程大学 | Network fault diagnosis method, equipment and storage medium based on wavelet neural network |
| US20220124543A1 (en) * | 2021-06-30 | 2022-04-21 | Oner Orhan | Graph neural network and reinforcement learning techniques for connection management |
-
2022
- 2022-05-18 CN CN202210542356.2A patent/CN114997217B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101762534B1 (en) * | 2016-02-23 | 2017-07-28 | 호남대학교 산학협력단 | System for remote diagnosing failure and monitoring of electric equipment using low power wide area network |
| CN110132596A (en) * | 2019-04-24 | 2019-08-16 | 昆明理工大学 | A method of the rolling bearing fault diagnosis based on wavelet packet and GWO-SVM |
| CN113125992A (en) * | 2021-04-23 | 2021-07-16 | 合肥工业大学 | NPC three-level inverter fault diagnosis method and system based on DBN |
| US20220124543A1 (en) * | 2021-06-30 | 2022-04-21 | Oner Orhan | Graph neural network and reinforcement learning techniques for connection management |
| CN113923104A (en) * | 2021-12-07 | 2022-01-11 | 南京信息工程大学 | Network fault diagnosis method, equipment and storage medium based on wavelet neural network |
Non-Patent Citations (3)
| Title |
|---|
| 何怡刚;汪涛;施天成;童晋;: "基于RFID传感器标签与深度学习的变压器状态监测方法研究", 电子测量与仪器学报, no. 09, 15 September 2018 (2018-09-15) * |
| 张朝龙;何怡刚;杜博伦;张兰芳;江善和;: "基于深度学习的电力变压器智能故障诊断方法", 电子测量与仪器学报, no. 01, 15 January 2020 (2020-01-15) * |
| 张鋆;张明皓;仝杰;李荡;雷煜卿;张树华;: "用于电力资产在线感知的eRFID标签设计", 电工技术学报, no. 11, 10 June 2020 (2020-06-10) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256592A (en) * | 2022-11-28 | 2023-06-13 | 国网山东省电力公司德州供电公司 | Medium-voltage distribution cable latent fault detection method and system |
| CN116256592B (en) * | 2022-11-28 | 2023-09-26 | 国网山东省电力公司德州供电公司 | Medium-voltage distribution cable latent fault detection method and system |
| CN117150374A (en) * | 2023-10-27 | 2023-12-01 | 国网江西省电力有限公司电力科学研究院 | Transformer fault identification method and system based on feature importance sorting |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114997217B (en) | 2024-09-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114997217B (en) | Transformer fault diagnosis method based on RFID sensing tag and IGWO-ELM | |
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| Bayati et al. | MLPSO: a filter multi-label feature selection based on particle swarm optimization | |
| CN111988329B (en) | Network intrusion detection method based on deep learning | |
| Chakraborty | Feature subset selection by particle swarm optimization with fuzzy fitness function | |
| CN110705760A (en) | Photovoltaic power generation power prediction method based on deep belief network | |
| CN110414788A (en) | A kind of power quality prediction technique based on similar day and improvement LSTM | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| WO2022010731A1 (en) | Compact representation and time series segment retrieval through deep learning | |
| Wu | A density adjustment based particle swarm optimization learning algorithm for neural network design | |
| CN116538127B (en) | Axial flow fan and control system thereof | |
| CN112149883A (en) | Photovoltaic power prediction method based on FWA-BP neural network | |
| CN108983180A (en) | A kind of high-precision radar sea clutter forecast system of colony intelligence | |
| Guan et al. | Application of a novel PNN evaluation algorithm to a greenhouse monitoring system | |
| CN112329370B (en) | Online fault diagnosis method for SiC power tube based on extreme learning machine | |
| Li et al. | Class imbalanced fault diagnosis via combining K-means clustering algorithm with generative adversarial networks | |
| Wu et al. | Fault diagnosis of TE process based on incremental learning | |
| Alanis et al. | A wind speed neural model with particle swarm optimization kalman learning | |
| CN115412332B (en) | Internet of things intrusion detection system and method based on hybrid neural network model optimization | |
| Jamsandekar et al. | Self generated fuzzy membership function using ANN clustering technique | |
| Li et al. | A short-term wind power forecasting method based on NWP wind speed fluctuation division and clustering | |
| CN115293249A (en) | Power system typical scene probability prediction method based on dynamic time sequence prediction | |
| CN114444763A (en) | Wind power prediction method based on AFSA-GNN | |
| Li et al. | Short-term Photovoltaic Power Forecasting Using SOM-based Regional Modelling Methods | |
| Wu et al. | Projective clustering using neural networks with adaptive delay and signal transmission loss |
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 |