CN109521299B - Intelligent fault reasoning method for inverter - Google Patents
Intelligent fault reasoning method for inverter Download PDFInfo
- Publication number
- CN109521299B CN109521299B CN201811426825.4A CN201811426825A CN109521299B CN 109521299 B CN109521299 B CN 109521299B CN 201811426825 A CN201811426825 A CN 201811426825A CN 109521299 B CN109521299 B CN 109521299B
- Authority
- CN
- China
- Prior art keywords
- fault
- probability
- inverter
- layer
- node
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Abstract
The invention discloses an intelligent inverter fault reasoning method, which is an intelligent inverter fault reasoning system of a three-layer Bayesian network consisting of an inverter operation condition layer, an inverter fault layer and a fault symptom layer. The invention has the advantages that: the inverter operation state is used as the first layer of the inference network, and compared with a common two-layer naive Bayesian network with independent variables, various complex faults can be inferred, the inference method better conforms to the inference thinking and the inference strategy of experts, embodies stronger intelligence, and can deal with various complex cause and effect relationships and uncertainties appearing at any time. Through example analysis of various evidences, the method can accurately reason out single faults under complete evidences, single faults under incomplete information and composite faults under incomplete information, can comprehensively judge the faults and the reasons thereof by combining equipment operation layer information and incomplete fault symptom information, shows stronger reasoning capability under the incomplete information, and has stronger practical guiding significance.
Description
Technical Field
The invention relates to the technical field of fault diagnosis in the power field, in particular to an intelligent fault reasoning method for an inverter.
Background
The existing inverter fault reasoning diagnosis method mainly comprises four categories, namely a method based on a neural network, an expert system method, a fault tree mode diagnosis method and a method based on comparison of detection quantity. The neural network based method acquires fault data through a sensor, and interprets encoded data of the sensor through the neural network. The expert system method is to establish a knowledge base and to judge the type of the fault by looking up the knowledge base through the phenomenon of the fault. The fault tree mode diagnosis method is to establish a fault tree and carry out fault diagnosis through an algorithm. The method based on the comparison of the detected quantity is to perform fault diagnosis by comparing the difference between the current or voltage in the normal and fault states of the inverter. Several of the above methods are capable of fault reasoning and diagnosis for inverters, but most of them require complete and complete fault signature information and fault data. However, it is very difficult to collect accurate and complete fault symptom information in practice, so that the diagnosis method has the problems of poor applicability, inaccurate diagnosis result and the like.
Disclosure of Invention
The invention aims to provide a reasoning and diagnosis method for inverter faults under the condition of incomplete information.
In order to achieve the purpose, the invention is implemented according to the following technical scheme: a method for intelligent fault reasoning of an inverter comprises the following steps:
the method comprises the following steps: establishing a three-layer Bayesian network structure
The first layer is an operation state layer of the inverter and consists of a plurality of inverter operation state variable nodes;
the second layer is an inverter fault layer and consists of a plurality of inverter fault nodes;
the third layer is an inverter fault symptom layer and consists of a plurality of inverter fault symptom nodes;
determining causal relationships among nodes of a first layer, a second layer and a third layer by field experts and field technicians, and constructing a three-layer Bayesian network structure;
in a bayesian network structure, each node represents a variable; edges represent the links between variables, pointing from parent nodes to leaf nodes, whose link strength can be represented by a conditional probability distribution; defining nodes without father nodes as root nodes, and defining other nodes as leaf nodes;
step two, determining prior probability and conditional probability of each node
The prior probability of the root node is obtained by adopting a historical operation data correction basic probability method; acquiring the conditional probability of the non-root node by adopting a prior Beta distribution method;
(ii) prior probability of root node
According to the pulse missing probability, I, of the inverterGBT open circuit probability and capacitance aging historical operation data, and setting the basic probability of the occurrence of the root node state as P0；
In the aspect of regular maintenance of a certain part of the inverter, if the regular maintenance is not carried out, the corresponding node correction value is increased, and the regular maintenance is not increased; in the aspect of the operation age of a certain part of the inverter, the correction value of the corresponding age of the corresponding node is increased; in the aspect of historical operating conditions of certain parts of the inverter, the correction value of corresponding times is increased every time when a fault occurs;
the formula of the historical operation data correction basic probability method of the prior probability of the root node is as shown in the formula (1);
where c is the number of factors that have an effect on the root node probability, PiIs the base probability of each influencing factor, aiFor each factor of influence, i ∈ [1, c ∈ ]](ii) a The data settings are shown in table 1;
table 1: root node prior probability profile
Assigning values to all root nodes of the inverter by the method and the formula (1) to determine a prior probability table of the root nodes;
(ii) conditional probability of non-root node
The probability of non-root nodes obeys normal distribution P (X) to (mu, sigma)2) Where P (X) is the probability value of the non-root node, μ is the expectation of a normal distribution, σ2Is the variance of a normal distribution; determining the node probability by using a prior Beta distribution method;
assuming that the number of the non-root nodes is h, a certain non-root node X is setrA state X ofrThe probability value interval [ mu-, mu + of +]Where the sum of μ is set manually based on historical data and expert experience, node XrHas mrA state value, k ∈ [1, m ]r](ii) a Obtaining parameters by calculating equation (2)Sigma; wherein r ∈ [1, h ]]Setting the upper and lower limits of probability values;
if node XrParent node Pa (X)r) Is combined with srThen set node XrHas a conditional probability of thetarjk＝P(Xr＝k|Pa(Xr)＝j)；
Existing sample data set D ═ D1,d2,...,dl) Defining a sample D in the data set De(e∈[1,l]) Characteristic function of
When X isrK and Pa (X)r) When the number j is equal to the number j,
other where j ∈ [1, s ]r]；
Where l is the number of data set samples, MrjkThe number of samples meeting the requirements in the data group is obtained;
the probability density function of the Beta distribution is:
the expectation of the Beta function is
Variance of
The expectation of normal distribution and Beta distribution are equal, the Beta distribution is a convex function as a precondition, the variance of the two is the closest to the target, and a functional relation is constructed, namely
Calculating formula (5) under the conditions that alpha is more than 1 and beta is more than 1;
calculating a Beta distribution parameter by using a formula (5), and obtaining prior distribution of a Beta function by using a formula (6);
π(θ)＝Be(arj1,arj2,...,arjk) (6)
determination of θ by equation (7)rjkBayesian estimate of (a);
θrjk＝(Mrjk+arjk)/(∑kMrjk+∑karjk) (7)
thirdly, acquiring a conditional probability table of the Bayesian network for all nodes according to the first step or the second step, and combining the conditional probability table with the Bayesian network structure established in the first step to form a complete Bayesian network;
step three, obtaining evidence information
Monitoring fault symptoms through a fault information acquisition system, taking acquired symptom information as an evidence, and setting an evidence set to be psi; inputting the acquired symptom information into a Bayesian network for fault reasoning;
step four, inverter fault reasoning
Inputting the collected fault symptom information into an algorithm model as reasoning evidence of a Bayesian network, and calculating a node X of a fault layer of an inverter under a given evidence psi by using a formula (8)vA state X ofvConditional probability of g:
wherein v ∈ [1, t ]]T is the number of fault layer nodes, node XvHas mvState, g ∈ [1, m ]v]ψ is a given set of evidence;
calculating fault layer of inverterAfter the probability of each node, the fault type of the inverter is distinguished by two rules. Rule 1: the fault with the highest probability is greater than a certain threshold1(ii) a Rule 2: the difference between the maximum failure probability and other failure probabilities is greater than a certain threshold2(ii) a If any rule is satisfied, the fault is judged to be a single fault, and if the rule is not satisfied, the fault is judged to be a composite fault;
determining the fault of the inverter according to the probability of each node of the fault layer; calculating the combined probability of each state of the inverter fault through an equation (9);
where ψ is a given set of evidence, Pa (X)v) Is combined with svW is equal to [1, s ]v]。Pa(Xv) Is XvIs a node variable of the inverter running state layer and is obtained by calculating Pa (X)v) W and XvAnd (5) positioning a first probability fault point according to the simultaneous probability of g, and then checking according to the fault probability from large to small until the fault point is checked.
Preferably, the inverter operation state variable nodes include phase-A upper bridge arm pulse loss, phase-C lower bridge arm pulse loss, phase-B upper bridge arm pulse loss, phase-A lower bridge arm pulse loss, phase-C upper bridge arm pulse loss, phase-B lower bridge arm pulse loss, phase-A upper bridge arm IGBT open circuit, phase-C lower bridge arm IGBT open circuit, phase-B upper bridge arm IGBT open circuit, phase-A lower bridge arm IGBT open circuit, phase-C upper bridge arm IGBT open circuit, phase-B lower bridge arm IGBT open circuit, upper capacitor C1Fault, lower capacitance C2Fault, capacitance parameter variation.
Preferably, the inverter fault nodes include an a-phase fault, a B-phase fault, a C-phase fault, and a DC link fault.
Preferably, the inverter fault symptom node includes a-phase positive waveform distortion, a-phase negative waveform distortion, B-phase positive waveform distortion, B-phase negative waveform distortion, C-phase positive waveform distortion, C-phase negative waveform distortion, voltage waveform abnormal pulse, upper capacitor C1Abnormal voltage, lower capacitance C2Abnormal voltage, upper capacitor C1And a lower capacitor C2The voltage waveform is asymmetric.
Preferably, P is the prior probability of the root node in the set-up table0＝0.05，P1＝0.05，P2＝0.02，P3＝0.1。
Preferably, the1＝80％，2＝20％。
Compared with the prior art, the invention has the advantages that: the invention provides an inverter intelligent fault reasoning system of a three-layer Bayesian network consisting of an inverter operation condition layer, an inverter fault layer and a fault symptom layer, wherein the inverter operation condition is used as the first layer of the reasoning network, and compared with a common two-layer naive Bayesian network with independent variables, the invention can infer various complex faults, better conforms to the reasoning idea and reasoning strategy of experts, embodies stronger intelligence and can cope with various complex cause and effect relationships and uncertainties appearing at any time. Through example analysis of various evidences, the method can accurately reason single faults under complete evidences, single faults under incomplete information and composite faults under incomplete information, can comprehensively judge the faults and the reasons thereof by combining equipment operation layer information and incomplete fault symptom information, shows stronger reasoning capability under the incomplete information, and has stronger practical significance.
Drawings
Fig. 1 is a three-layer bayesian network structure diagram of the intelligent fault inference of the inverter in embodiment 1 or 2 of the present invention.
Fig. 2 is a flowchart of intelligent inverter fault inference according to embodiment 1 or 2 of the present invention.
Fig. 3 is a flowchart for constructing an intelligent fault inference bayesian network for an inverter according to embodiment 1 or 2 of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and specific embodiments, which are illustrative of the invention and are not to be construed as limiting the invention.
An intelligent inverter fault reasoning system based on a three-layer Bayesian network is characterized in that firstly, a Bayesian network framework structure graph model shown in FIG. 1 is established according to actual diagnosis processes and ideas of field technicians by combining field expert knowledge; secondly, determining variables of each layer, establishing a Bayesian network according to the causal relationship of the variables, and determining a conditional probability table of the network by using a historical operation data correction basic probability method and a prior Beta distribution method; thirdly, the fault information acquisition system acquires fault symptom information, converts the fault symptom information into evidence information, inputs the evidence information into a Bayesian network, and calculates probability data of a fault layer of the inverter through a Bayesian algorithm; then, judging the fault type through rule differentiation, and determining the fault of the inverter through the probability of each node of a fault layer; and finally, determining a first probability fault point by calculating the maximum state combination, and then checking according to the fault probability from large to small until the fault point is checked, wherein an intelligent fault reasoning flow chart of the inverter is shown in fig. 2.
In this example, the data is from maintenance records, inspection logs and test data of a frequency converter (model: ACS6000) of a hoisting machine of a main shaft transmission system of Zhao Gu II mine.
The method provided by the invention is used for fault diagnosis of the hoist inverter, and a Bayesian network construction flow chart is shown in FIG. 3, and the method comprises the following specific steps:
the first step is as follows: and determining variables and causal relationships among the variables by combining knowledge of field experts and field technicians and historical operating data of the inverter to establish a three-layer Bayesian network framework structure. The structure is divided into three layers, wherein variables listed in the first layer are various faults which directly cause the frequency converter to appear; variables listed in the second tier are various inverter faults; the variables listed in the third level are symptoms of various faults in the inverter. And determining the causal relationship among the variables, and connecting the nodes with the causal relationship by using directed arrows to construct a three-layer Bayesian network structure.
The second step is that: the probability of the root node is determined using a corrected base probability method based on historical operating data in conjunction with historical data of inverter operation. Firstly, setting the root node basic probability to be 0.05, and correcting the node probability according to the three aspects of whether the inverter is maintained regularly, the running life of the inverter and the failure frequency of the inverter. In the example, the basic probability is increased by 0.05 if the equipment is not periodically overhauled; the basic probability of the equipment is increased by 0.02 every year of operation, and the basic probability is increased by 0.1 every time the equipment fails.
As shown in FIG. 1, node X1The represented variable is the loss of the pulse of the upper bridge arm of the phase A, and the historical operating conditions of the upper bridge arm of the phase A are as follows: periodically overhauled for 1 year t11 is ═ 1; no fault occurred, t20. Then formula 1 is substituted:
wherein a is1＝0；a2＝t1；a3＝t2。
Calculating to obtain P (X)1)＝0.07。
And the other root nodes respectively determine the root node probability according to the method according to the conditions of the negative period overhaul, the inverter operation age and the inverter failure frequency, so as to obtain all root node probability sets.
The third step: setting the value intervals of the non-root nodes according to historical data and expert knowledge, setting mu to be 0.3 and mu to be 0.2, and calculating through an equation (2) to obtain a parameter sigma.
Solved to be 0.05.
The fourth step: according to the existing sample data set D ═ D1,d2,...,dl) Calculating the number M of parameters meeting the requirements in the data set by the formula (3)rjk。
In the following embodiments, each node variable has two states: occurrence and non-occurrence. The occurrence of the state is represented by 2, and the absence of the state is represented by 1.
For example statistically in Pa (X)16) Calculating X under the condition of 116The bayesian parameters for the data set sample at 1 are:
wherein s is16＝16。
The fifth step: parameters α and β of the Beta distribution are obtained by approximating the normal distribution using the Beta distribution. In that
Alpha > 1 and beta > 1:
the Beta distribution function Beta (alpha, Beta) is obtained.
In that
Beta distribution functions Beta (73,1) were calculated under the condition that σ was 0.05.
And a sixth step: taking prior distribution pi (theta) of Beta function as Be (a)rj1,arj2,...,arjk) (6)
And (3) under the data of the fifth step, substituting formula 6 into the prior distribution of the Beta function, wherein pi (theta) is equal to Be (73, 1).
The seventh step: calculated to obtain thetarjkBayesian estimate of (a). The calculation formula is as shown in formula (7):
θrjk＝(Mrjk+arjk)/(∑kMrjk+∑karjk) (7)
calculating to obtain theta by using the data from the third step to the seventh step16,1,1(25+73)/(26+74) 0.98. Wherein
Eighth step: and repeating the third step to the seventh step to obtain the conditional probabilities of all the non-root nodes, and combining the Bayesian network structures obtained in the first step and the second step and the prior probabilities of the root nodes to obtain a complete Bayesian network.
And (3) carrying out fault reasoning under the condition of insufficient evidence information and conflict evidence information by using the Bayesian network obtained by the method:
and the fault type of the inverter is distinguished by two rules: rule 1: the fault with the highest probability is greater than a certain threshold1. Rule 2: the difference between the maximum failure probability and other failure probabilities is greater than a certain threshold2. In the following examples, the maintenance record, test data and expert knowledge consideration are combined and set1＝80％，220% by weight. Satisfaction of any one rule may be determined as a single failure. And the first rule and the second rule are not satisfied, and the fault type is judged to be a composite fault.
Example 1 evidence information is insufficient
With reference to fig. 1, the symptom information collected by the fault information collection system includes: b-phase negative waveform distortion generation (X)232) and C-phase positive waveform distortion generation (X)24＝2)。
The single failure probability data of the failure layer calculated by combining equation (8) is:
P(X16＝2|X23＝2,X24＝2)＝15.02％
P(X17＝2|X23＝2,X24＝2)＝25.29％
P(X18＝2|X23＝2,X24＝2)＝32.25％
P(X19＝2|X23＝2,X24＝2)＝52.87％
the probability of DC link fault occurrence is maximum, then the single fault probability is regularly distinguished, P (X)19＝2|X23＝2,X24＝2)-P(X18＝2|X23＝2,X242-20.62% > 20%, the case satisfies rule 2, the fault type is determined as single fault, and the inverter fault can be determined as DC link fault (X)19)。
The parent node of the DC link fault is as follows: upper capacitor fault (X)13) Lower capacitor fault (X)14) And change in capacitance parameter (X)15) Calculating each father of the DC link fault through a formula (9)The probability of node concurrence is obtained:
P(X13＝2,X19＝2|X23＝2,X24＝2)＝17.35％
P(X14＝2,X19＝2|X23＝2,X24＝2)＝19.69％
P(X15＝2,X19＝2|X23＝2,X24＝2)＝20.64％
determining a first probability fault point as a change (X) in the capacitance parameter from the calculated result15) And (4) according to the actual situation.
Example 2 evidence information Conflict
With reference to fig. 1, the symptom information collected by the fault information collection system includes: a-phase negative waveform distortion generation (X)212) and no B-phase positive waveform distortion (X)211) and B-phase negative waveform distortion generation (X)232), and the occurrence of the loss of the pulse of the lower bridge arm of the B phase is found by inspection (X)62). From the evidence given, this case is the case of evidence conflict. Evidence comes from the inverter health level and the fault symptom level. Evidence from run-time layer B phase lower leg pulse absence (X)6) Indicating the presence of a phase B fault (X)17) But this conclusion is not supported by the failing symptom layer.
The single failure probability data of the failure layer calculated by combining equation (8) is:
P(X16＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝24.51％
P(X17＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝12.75％
P(X18＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝48.87％
P(X19＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝42.55％
and C-phase faults have the maximum probability, then the single fault probability is subjected to rule distinguishing, the rule I and the rule II are not satisfied, and the fault type of the case is judged as a composite fault. Calculating the probability of the C-phase fault and the simultaneous occurrence of each father node by the formula (9) to obtain:
P(X2＝2,X18＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝9.76％
P(X5＝2,X18＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝48.87％
P(X8＝2,X18＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝4.56％
P(X11＝2,X18＝2|X6＝1,X21＝2,X22＝2,X23＝2)＝13.06％
calculating the probability of simultaneous occurrence of each father node of the DC link fault to obtain:
P(X13＝2,X19＝2|X6＝2,X21＝2,X22＝2,X23＝2)＝13.96％
P(X14＝2,X19＝2|X6＝2,X21＝2,X22＝2,X23＝2)＝15.85％
P(X15＝2,X19＝2|X6＝2,X21＝2,X22＝2,X23＝2)＝16.61％
determining the probability fault point as C-phase upper bridge arm pulse missing (X) according to the calculated result5) And change in capacitance parameter (X)15) And (4) according to the actual situation.
The reasoning results of the two embodiments are consistent with the actual situation, and the method provided by the invention is effective and accurate.
The technical solution of the present invention is not limited to the limitations of the above specific embodiments, and all technical modifications made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims (6)
1. An intelligent fault reasoning method for an inverter is characterized in that: the method comprises the following steps:
the method comprises the following steps: establishing a three-layer Bayesian network structure
The first layer is an operation state layer of the inverter and consists of a plurality of inverter operation state variable nodes;
the second layer is an inverter fault layer and consists of a plurality of inverter fault nodes;
the third layer is an inverter fault symptom layer and consists of a plurality of inverter fault symptom nodes;
determining causal relationships among nodes of a first layer, a second layer and a third layer by field experts and field technicians, and constructing a three-layer Bayesian network structure;
in a bayesian network structure, each node represents a variable; edges represent the links between variables, pointing from parent nodes to leaf nodes, whose link strength can be represented by a conditional probability distribution; defining nodes without father nodes as root nodes, and defining other nodes as leaf nodes;
step two, determining prior probability and conditional probability of each node
The prior probability of the root node is obtained by adopting a historical operation data correction basic probability method; acquiring the conditional probability of the non-root node by adopting a prior Beta distribution method;
(ii) prior probability of root node
Setting the basic probability of the occurrence of the root node state as P according to the pulse missing probability of the inverter, the IGBT open-circuit probability and the historical operation data of capacitor aging0；
In the aspect of regular maintenance of a certain part of the inverter, if the regular maintenance is not carried out, the corresponding node correction value is increased, and the regular maintenance is not increased; in the aspect of the operation age of a certain part of the inverter, the correction value of the corresponding age of the corresponding node is increased; in the aspect of historical operating conditions of certain parts of the inverter, the correction value of corresponding times is increased every time when a fault occurs;
the formula of the historical operation data correction basic probability method of the prior probability of the root node is as shown in the formula (1);
where c is the number of factors that have an effect on the root node probability, PiIs the base probability of each influencing factor, aiFor each factor of influence, i ∈ [1, c ∈ ]](ii) a The data settings are shown in table 1;
table 1: root node prior probability profile
Assigning values to all root nodes of the inverter by the method and the formula (1) to determine a prior probability table of the root nodes;
(ii) conditional probability of non-root node
The probability of non-root nodes obeys normal distribution P (X) to (mu, sigma)2) Where P (X) is the probability value of the non-root node, μ is the expectation of a normal distribution, σ2Is the variance of a normal distribution; determining the node probability by using a prior Beta distribution method;
assuming that the number of the non-root nodes is h, a certain non-root node X is setrA state X ofrThe probability value interval [ mu-, mu + of +]Where the sum of μ is set manually based on historical data and expert experience, node XrHas mrA state value, k ∈ [1, m ]r](ii) a Obtaining a parameter sigma by calculating the formula (2); wherein r ∈ [1, h ]]Setting the upper and lower limits of probability values;
if node XrParent node Pa (X)r) Is combined with srThen set node XrHas a conditional probability of thetarjk＝P(Xr＝k|Pa(Xr)＝j)；
Existing sample data set D ═ D1,d2,...,dl) Defining a sample D in the data set DeCharacteristic function of
When X isrK and Pa (X)r) When the number j is equal to the number j,
wherein e ∈ [1, l ]]，j∈[1,sr]；
Where l is the number of data set samples, MrjkThe number of samples meeting the requirements in the data group is obtained;
the probability density function of the Beta distribution is:
the expectation of the Beta function is
Variance of
The expectation of normal distribution and Beta distribution are equal, the Beta distribution is a convex function as a precondition, the variance of the two is the closest to the target, and a functional relation is constructed, namely
α>1,β>1, calculating formula (5);
calculating a Beta distribution parameter by using a formula (5), and obtaining prior distribution of a Beta function by using a formula (6);
π(θ)＝Be(arj1,arj2,...,arjk) (6)
determination of θ by equation (7)rjkBayesian estimate of (a);
θrjk＝(Mrjk+arjk)/(∑kMrjk+∑karjk) (7)
thirdly, acquiring a conditional probability table of the Bayesian network for all nodes according to the first step or the second step, and combining the conditional probability table with the Bayesian network structure established in the first step to form a complete Bayesian network;
step three, obtaining evidence information
Monitoring fault symptoms through a fault information acquisition system, taking acquired symptom information as an evidence, and setting an evidence set to be psi; inputting the acquired symptom information into a Bayesian network for fault reasoning;
step four, inverter fault reasoning
Inputting the collected fault symptom information into an algorithm model as reasoning evidence of a Bayesian network, and calculating a node X of a fault layer of an inverter under a given evidence psi by using a formula (8)vA state X ofvConditional probability of g:
wherein v ∈ [1, t ]]T is the number of fault layer nodes, node XvHas mvState, g ∈ [1, m ]v]ψ is a given set of evidence;
after the probability of each node of the inverter fault layer is calculated, the fault type of the inverter is distinguished through two rules; rule 1: the fault with the highest probability is greater than a certain threshold1(ii) a Rule 2: the difference between the maximum failure probability and other failure probabilities is greater than a certain threshold2(ii) a If any rule is satisfied, the fault is judged to be a single fault, and if the rule is not satisfied, the fault is judged to be a composite fault;
determining the fault of the inverter according to the probability of each node of the fault layer; calculating the combined probability of each state of the inverter fault through an equation (9);
where ψ is a given set of evidence, Pa (X)v) Is combined with svW is equal to [1, s ]v]；Pa(Xv) Is XvIs a node variable of the inverter running state layer and is obtained by calculating Pa (X)v) W and XvAnd (5) positioning a first probability fault point according to the simultaneous probability of g, and then checking according to the fault probability from large to small until the fault point is checked.
2. The method for intelligent inverter fault inference according to claim 1, wherein: the inverter running state variable node comprises an A-phase upper bridge arm pulse loss, a C-phase lower bridge arm pulse loss, a B-phase upper bridge arm pulse loss, an A-phase lower bridge arm pulse loss, a C-phase upper bridge arm pulse loss, a B-phase lower bridge arm pulse loss, an A-phase upper bridge arm IGBT open circuit, a C-phase lower bridge arm IGBT open circuit, a B-phase upper bridge arm IGBT open circuit, an A-phase lower bridge arm IGBT open circuit, a C-phase upper bridge arm IGBT open circuit, a B-phase lower bridge arm IGBT open circuit, an upper capacitor C1Fault, lower capacitance C2Faults and capacitance parameter variations.
3. The method for intelligent inverter fault inference according to claim 1, wherein: the inverter fault nodes include an A-phase fault, a B-phase fault, a C-phase fault and a DC link fault.
4. The method for intelligent inverter fault inference according to claim 1, wherein: the inverter fault symptom node comprises A-phase positive waveform distortion, A-phase negative waveform distortion, B-phase positive waveform distortion, B-phase negative waveform distortion, C-phase positive waveform distortion, C-phase negative waveform distortion, voltage waveform abnormal pulse and upper capacitor C1Abnormal voltage, lower capacitance C2Voltage anomaly and upper capacitance C1And a lower capacitor C2The voltage waveform is asymmetric.
5. An inverter according to claim 1A method of fault reasoning, characterized by: in the prior probability table of root node, P0＝0.05，P1＝0.05，P2＝0.02，P3＝0.1。
6. The method for intelligent inverter fault inference according to claim 1, wherein: the above-mentioned1＝80％，2＝20％。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811426825.4A CN109521299B (en) | 2018-11-27 | 2018-11-27 | Intelligent fault reasoning method for inverter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811426825.4A CN109521299B (en) | 2018-11-27 | 2018-11-27 | Intelligent fault reasoning method for inverter |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109521299A CN109521299A (en) | 2019-03-26 |
| CN109521299B true CN109521299B (en) | 2020-11-03 |
Family
ID=65793555
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811426825.4A Active CN109521299B (en) | 2018-11-27 | 2018-11-27 | Intelligent fault reasoning method for inverter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109521299B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005055190A (en) * | 2003-08-01 | 2005-03-03 | Japan Science & Technology Agency | Electromagnetic-wave radiation source detecting method and device by bayesian network |
| CN101915234A (en) * | 2010-07-16 | 2010-12-15 | 西安交通大学 | Method for diagnosing compressor-associated failure based on Bayesian network |
| CN102262690A (en) * | 2011-06-07 | 2011-11-30 | 中国石油大学(北京) | Modeling method of early warning model of mixed failures and early warning model of mixed failures |
| CN103729553A (en) * | 2013-12-19 | 2014-04-16 | 浙江工商大学 | Classification control method for urban safety complex events on basis of Bayesian network learning |
| CN104698397A (en) * | 2015-03-16 | 2015-06-10 | 浙江万里学院 | Fault diagnosis method of multi-level inverter |
| CN105608634A (en) * | 2015-11-09 | 2016-05-25 | 国网新疆电力公司 | Bayesian network based electrical network risk early-warning evaluation model |
| CN108051660A (en) * | 2017-10-31 | 2018-05-18 | 华北电力大学（保定） | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method |
| CN108267684A (en) * | 2018-01-15 | 2018-07-10 | 哈尔滨理工大学 | A kind of Converter Fault Diagnosis method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050021237A1 (en) * | 2003-07-03 | 2005-01-27 | Schachter Asher Daniel | Method and apparatus for evaluating new chemical entities |
-
2018
- 2018-11-27 CN CN201811426825.4A patent/CN109521299B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2005055190A (en) * | 2003-08-01 | 2005-03-03 | Japan Science & Technology Agency | Electromagnetic-wave radiation source detecting method and device by bayesian network |
| CN101915234A (en) * | 2010-07-16 | 2010-12-15 | 西安交通大学 | Method for diagnosing compressor-associated failure based on Bayesian network |
| CN102262690A (en) * | 2011-06-07 | 2011-11-30 | 中国石油大学(北京) | Modeling method of early warning model of mixed failures and early warning model of mixed failures |
| CN103729553A (en) * | 2013-12-19 | 2014-04-16 | 浙江工商大学 | Classification control method for urban safety complex events on basis of Bayesian network learning |
| CN104698397A (en) * | 2015-03-16 | 2015-06-10 | 浙江万里学院 | Fault diagnosis method of multi-level inverter |
| CN105608634A (en) * | 2015-11-09 | 2016-05-25 | 国网新疆电力公司 | Bayesian network based electrical network risk early-warning evaluation model |
| CN108051660A (en) * | 2017-10-31 | 2018-05-18 | 华北电力大学（保定） | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method |
| CN108267684A (en) * | 2018-01-15 | 2018-07-10 | 哈尔滨理工大学 | A kind of Converter Fault Diagnosis method |
Non-Patent Citations (1)
| Title |
|---|
| 基于 BP 神经网络的逆变器开路故障诊断方法;韩素敏等;《河南理工大学学报（自然科学版）》;20180930;第37卷(第5期);第122-127页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109521299A (en) | 2019-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086889B (en) | Terminal fault diagnosis method, device and system based on neural network | |
| CN111047082B (en) | Early warning method and device of equipment, storage medium and electronic device | |
| Schlechtingen et al. | Wind turbine condition monitoring based on SCADA data using normal behavior models. Part 2: Application examples | |
| US20110020122A1 (en) | Integrated condition based maintenance system for wind turbines | |
| CN108335021A (en) | A kind of method and maintenance decision optimization of wind energy conversion system state Reliability assessment | |
| CN111103137A (en) | Wind turbine gearbox fault diagnosis method based on deep neural network | |
| CN111506049B (en) | Multiple fault diagnosis method for aero-engine control system based on AANN network system | |
| KR102545672B1 (en) | Method and apparatus for machine fault diagnosis | |
| Abbasi et al. | Development of predictive maintenance interface using multiple linear regression | |
| CN115081647A (en) | Industrial intelligent instrument fault pre-diagnosis method based on Bayesian network model | |
| CN109521299B (en) | Intelligent fault reasoning method for inverter | |
| JP7191807B2 (en) | Diagnostic device and diagnostic method | |
| US11339763B2 (en) | Method for windmill farm monitoring | |
| CN111413564A (en) | Supercapacitor failure early warning method, system and equipment | |
| CN115822894A (en) | Operation and maintenance method and system for offshore wind power equipment | |
| CN110083804B (en) | Wind power plant SCADA data missing intelligent repairing method based on condition distribution regression | |
| Zhang | Comparison of data-driven and model-based methodologies of wind turbine fault detection with SCADA data | |
| Souza et al. | Evaluation of data based normal behavior models for fault detection in wind turbines | |
| CN111986469A (en) | Intelligent diagnosis method for field terminal fault | |
| CN114046228B (en) | Wind turbine generator abnormality diagnosis method and system | |
| RU2546993C1 (en) | Method to diagnose technical condition of electric drive by evaluation of dynamics of its electrical parameters | |
| CN117741429A (en) | Method, device, medium and equipment for determining health index of ship alternating current motor | |
| WO2023102630A1 (en) | Method for identifying anomalies and predicting failures and processor-readable storage means | |
| CN117150360A (en) | Wind turbine generator system fault tracing method and system based on SCADA parameter correlation and DNN | |
| CN116451121A (en) | Switch fault diagnosis method, system, equipment and medium |
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 |