CN110119713A - Mechanical Failure of HV Circuit Breaker diagnostic method based on D-S evidence theory - Google Patents

Abstract

The present application discloses a method for diagnosing mechanical faults of high-voltage circuit breakers based on D-S evidence theory. The method includes: establishing an adaptive neuro-fuzzy reasoning system model; acquiring sound signals and vibration signals of high-voltage circuit breakers; extracting the sound The feature quantity of the signal and the vibration signal; through the self-adaptive neuro-fuzzy inference system model, the feature quantity is classified and preprocessed respectively to obtain the basic probability distribution function; the D-S evidence theory modified by high conflict evidence The probability distribution function is fused to obtain the mechanical fault diagnosis result. The diagnostic method provided by this application combines sound and vibration signals, uses D-S evidence theory for information fusion, and comprehensively diagnoses mechanical failures of high-voltage circuit breakers, improving the reliability and accuracy of state assessment.

Description

A Diagnosis Method for Mechanical Faults of High Voltage Circuit Breakers Based on D-S Evidence Theory

technical field

The present application relates to the technical field of high voltage circuit breaker mechanical fault diagnosis, in particular to a high voltage circuit breaker mechanical fault diagnosis method based on D-S evidence theory.

Background technique

Due to the rapid development of industrial automation, automated production lines have put forward higher requirements for power supply reliability and power quality. High voltage circuit breaker is one of the most important equipment in the whole power system, and its stable operation is a necessary condition for improving power supply reliability and power quality. Among the various failures of high-voltage circuit breakers, the main factors affecting the reliability of their actions during mechanical failures account for 70%-80% of the total failures. Therefore, it is necessary to monitor the mechanical characteristics of high-voltage circuit breakers online.

At present, the method of signal comparison is mainly used to monitor the mechanical characteristics of circuit breakers. When a certain characteristic signal changes from the normal situation, it indicates that a fault may have occurred. However, due to the complex operating state of the diagnostic object and many influencing factors, the same fault often manifests differently, and the same symptom may be a variety of faults. The fault characteristic quantity obtained by single signal detection generally cannot effectively complete the fault diagnosis, resulting in fault diagnosis. Accuracy is difficult to guarantee.

Contents of the invention

The present application provides a high-voltage circuit breaker mechanical fault diagnosis method based on D-S evidence theory to solve the problem of low accuracy of the current high-voltage circuit breaker mechanical fault diagnosis method.

In order to solve the above technical problems, the embodiment of the present application discloses the following technical solutions:

The embodiment of the present application discloses a method for diagnosing a mechanical fault of a high-voltage circuit breaker based on the D-S evidence theory. The method includes:

Build an adaptive neuro-fuzzy reasoning system model;

Obtain the sound signal and vibration signal of the high voltage circuit breaker;

Extracting the feature quantities of the sound signal and the vibration signal;

Perform classification preprocessing on the feature quantities respectively through the adaptive neuro-fuzzy inference system model to obtain a basic probability distribution function;

The basic probability distribution function is fused by the D-S evidence theory modified by high conflict evidence to obtain the mechanical fault diagnosis result.

Optionally, build an adaptive neuro-fuzzy inference system model, including:

Acquiring historical data of sound and vibration signals of the high-voltage circuit breaker;

Extracting the feature quantity of the historical data of the sound and vibration signal;

Adaptive neuro-fuzzy control training is performed on the feature quantities respectively, and an adaptive neuro-fuzzy reasoning system model is established.

Optionally, the network structure of the adaptive neuro-fuzzy inference system model includes a fuzzification layer, a rule inference layer, a normalization layer, a defuzzification layer and an output layer, wherein,

The fuzzification layer performs fuzzification processing on the input feature quantity, and the output function of node i is as follows:

In formula (1), x is the input of node i, A _i and B _i-2 are fuzzy sets, O _1,i is the membership function value of the fuzzy set, which reflects the degree to which x belongs to A _i , μA _i and μB _i-2 is the membership function of the fuzzy set;

In formulas (2) and (3), c _i and σ _i are the parameters of the membership function;

The rule reasoning layer calculates the incentive degree of each rule, and its output expression is:

In formula (4), w _i is the incentive strength of the corresponding rule, that is, the weight of each fuzzy rule;

The normalization layer calculates the ratio of the incentive strength of the i-th rule to the sum of the incentive strengths of all rules, and its output expression is:

In formula (5), is the normalized incentive strength of the i-th rule, indicating the contribution of the i-th rule to the final result;

The defuzzification layer calculates the output of each rule, and its output expression is:

In formula (6), the parameters p _i , q _i , r _i are the subsequent parameters of each node;

The output layer calculates the sum of all defuzzification node outputs and outputs,

In formula (7), Y is the total output of the system, and y _i is the output of the i-th rule.

Optionally, the basic probability distribution function is fused with the D-S evidence theory modified by highly conflicting evidence to obtain a mechanical fault diagnosis result, including:

Calculate the evidence distance d and the normalization constant k between the evidences;

averaging the evidence distance d and the normalization constant k to obtain the conflict coefficient σ;

Judging whether there is conflict evidence according to the conflict coefficient σ;

If there is conflicting evidence, modify the basic probability distribution function, and fuse the modified basic probability distribution function through D-S evidence theory;

If there is no conflicting evidence, the basic probability distribution function is fused directly through the D-S evidence theory.

Optionally, calculate the evidence distance d and the normalization constant k between the evidences, including:

The evidence distance d between evidence m ₁ and m ₂ is:

In formulas (8) and (9), A _i , B _j ∈ U, m ₁ (A _i ), m ₂ (B _j ) are the basic probability assignments of elements A and B respectively, and |A| number;

The formula for calculating the normalization constant k is:

In formula (10), k reflects the degree of conflict of evidence, and its value range is [0,1].

Optionally, if there is conflicting evidence, modifying the basic probability assignment function includes:

Compute the confidence for each evidence:

In formula (11), sim(m _i ,m _j ) is the similarity between evidence m _i and m _j , sim(m _i ,m _j )=1-d(m _i ,m _j );

Under the unified identification framework, conflicts among multiple sources of evidence are counted:

The conflict of the remaining evidence after removing the jth evidence source from the evidence sources is:

According to the conflict k ₀ and conflict k _j , the degree of falsity of the evidence is calculated:

According to the degree of trust and the degree of falsity, the weight of the evidence focal element distribution is calculated as:

τ _i =CrdPm _i -γ*Falm _i +1 (15)

In formula (15), if the conflict coefficient σ is greater than 0.5, then γ takes 1 or 2; if the conflict coefficient σ is greater than 0.9, then γ takes

The weight T _i of the conflict evidence is obtained after normalizing the weights:

The basic probability distribution function is modified according to the weight T _i .

Optionally, the rules for fusing the basic probability distribution function through the D-S evidence theory are:

in,

The beneficial effects of the high-voltage circuit breaker mechanical fault diagnosis method based on the D-S evidence theory provided by this application are as follows:

(1) The traditional mechanical failure diagnosis of high-voltage circuit breakers uses a single signal for monitoring and diagnosis, which has various disadvantages. This method combines sound and vibration signals to comprehensively diagnose the mechanical failure of high-voltage circuit breakers, which is more reliable and accurate;

(2) After the adaptive neuro-fuzzy inference system (ANFIS) is constructed in this method, it is no longer necessary to repeatedly extract and train historical data, and the speed of diagnosis can be improved under real-time monitoring conditions;

(3) This method uses the D-S evidence theory for information fusion. Aiming at the problem that the fusion result is inconsistent with the facts when there is high conflict evidence, a correction of high conflict evidence is proposed. On the basis of not changing the trust degree of the original evidence, only the conflict The correction of the evidence will not only not destroy the credibility of the source evidence, but also improve the convergence speed and accuracy to a certain extent, and the fusion result will not be contrary to the facts.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Description of drawings

In order to illustrate the technical solution of the present application more clearly, the accompanying drawings that need to be used in the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, on the premise of not paying creative work, there are also Additional figures can be derived from these figures.

Fig. 1 is a flow chart of a method for diagnosing a mechanical fault of a high-voltage circuit breaker based on the D-S evidence theory provided by an embodiment of the present application;

FIG. 2 is a detailed flow chart of S500 in the method for diagnosing a mechanical fault of a high voltage circuit breaker based on the D-S evidence theory provided by the embodiment of the present application.

Detailed ways

In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described The embodiments are only some of the embodiments of the present application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

The sound and vibration signals of high-voltage circuit breakers are non-stationary and non-deterministic signals, which can reflect the complex mechanical action process of circuit breakers. There are various disadvantages in fault diagnosis of single signal, and multi-signal fusion technology using D-S evidence theory can solve these disadvantages to a certain extent, and has great advantages in identifying circuit breaker fault types. Referring to FIG. 1 , it is a flow chart of a method for diagnosing a mechanical fault of a high-voltage circuit breaker based on the D-S evidence theory provided by an embodiment of the present application.

As shown in Figure 1, the method for diagnosing mechanical faults of high-voltage circuit breakers based on the D-S evidence theory provided by the embodiment of the present application includes:

S100: Establishing an adaptive neuro-fuzzy reasoning system model.

First, sort out the historical data of sound and vibration collected by the high-voltage circuit breaker, and select four states of normal, shaft jamming, base loosening, and opening refusal to diagnose and classify. Analyze to suppress Gaussian noise, and then perform Hilbert-Huang transform to obtain IMF energy entropy, which is used as the feature quantity for ANFIS (AdaptiveNetwork-based Fuzzy Inference System, Adaptive Neuro-Fuzzy System) training. Multi-resolution analysis is included, and then wavelet packet reconstruction is performed on the decomposition node coefficients of each layer on this basis. Finally, the phase space state matrix is reconstructed and the singular spectral entropy of the feature matrix is calculated as the feature quantity for ANFIS training.

The feature quantity of each 20 groups of processed samples is used as input and the number of training is selected as 200 times, and the adaptive neuro-fuzzy reasoning system is trained, and the error curve after the training is observed. The error is reduced to less than 0.05, indicating that the fuzzy control rule is available. , get the adaptive neuro-fuzzy inference system model of sound and vibration signal respectively.

The network structure of the adaptive neuro-fuzzy inference system model (taking double input and single output as an example) is as follows:

Layer 1: Fuzzy layer, used to fuzzy the input data, the output function of node i is as follows:

In formula (1), x is the input of node i, A _i and B _i-2 are fuzzy sets, O _1,i is the membership function value of the fuzzy set, which reflects the degree to which x belongs to A _i , μA _i and μB _i-2 is the membership function of the fuzzy set, and Gaussian function can be selected.

In formulas (2) and (3), c _i and σ _i are the parameters of the membership function, also known as the antecedent parameters.

Layer 2: rule reasoning layer, this layer calculates the incentive degree of each rule, and its output expression is:

O _2,i =w _i =μ _Ai (x ₁ )μ _Bi (x ₂ ),I=1,2 (4)

In formula (4), w _i is the incentive strength of the corresponding rule, that is, the weight of each fuzzy rule.

Layer 3: Normalization layer, which normalizes the incentive strength of each rule, that is, calculates the ratio of the incentive strength of the i-th rule to the sum of the incentive strength of all rules, and its output expression is:

In formula (5), is the normalized incentive strength of the i-th rule, indicating the contribution of the i-th rule to the final result.

Layer 4: Defuzzification layer, this layer calculates the output of each rule, and its output expression is:

In formula (6), the parameters p _i , q _i , r _i are the subsequent parameters of each node in this layer, which are obtained by ANFIS.

Layer 5: Output layer, this layer calculates the sum of all defuzzification node outputs and produces the final ANFIS output:

In formula (7), Y is the total output of the system, and y _i is the output of the i-th rule.

The hybrid algorithm of homoregressive gradient descent method and least squares method is used to identify the antecedent parameters and subsequent parameters of the system, and then realize the establishment of fuzzy model. For the hybrid algorithm, the learning process of each cycle includes two parts: forward transmission and back propagation. In the process of forward learning, the antecedent parameters are fixed, and the least square method is used to identify the latter parameters; in the process of reverse learning , calculate the error of the calculated subsequent parameters, adopt the BP algorithm in the feed-forward neural network, reversely transmit the error from the output end to the input end, and finally use the gradient descent method to update the antecedent parameters.

S200: Obtain the sound signal and vibration signal of the high voltage circuit breaker.

S300: Extract feature quantities of the sound signal and the vibration signal.

S400: Classifying and preprocessing the feature quantities through the adaptive neuro-fuzzy inference system model to obtain a basic probability distribution function.

S500: Fusing the basic probability distribution function with the D-S evidence theory modified by highly conflicting evidence to obtain a mechanical fault diagnosis result.

The sound signal and vibration signal of the high-voltage circuit breaker are collected, and after the sound signal is denoised, an appropriate method is used to extract the feature quantity of the sound and vibration signal to represent the mechanical feature information contained in the signal.

Taking the feature quantities of the sound signal and vibration signal as input, the feature quantities are classified and preprocessed through the established adaptive neuro-fuzzy inference system model, and the basic probability assignment function (BPA, Basic Probability Assignment) is obtained. The obtained basic probability assignment function The D-S evidence theory modified by high conflict evidence is used for fusion, and finally the mechanical fault diagnosis of high voltage circuit breaker based on sound and vibration signals is obtained.

The steps of multi-information fusion of D-S evidence theory based on high-conflict evidence correction are shown in Figure 2:

S501: Calculate the evidence distance d and the normalization constant k between the evidences:

Under the recognition framework U, if the function m: 2 ^u -> [0,1] satisfies the condition and Then m(A) is called the basic probability assignment of element A, and also represents the degree of trust in evidence A, where A that makes m(A)>0 is called the focal element. The evidence distance d between evidence m ₁ and m ₂ is:

In formulas (8) and (9), A _i , B _j ∈ U, |A| represent the number of contained elements, and the similarity between evidence m ₁ and m ₂ sim(m ₁ ,m ₂ )=1- d(m ₁ ,m ₂ ), can be extended to multiple evidences and represented by a similarity matrix. The larger the value, the greater the similarity between evidences and the smaller the conflict.

The formula for calculating the normalization constant k is:

In formula (10), k reflects the degree of conflict of evidence, and its value range is [0,1].

S502: Average the evidence distance d and the normalization constant k to obtain the conflict coefficient σ.

S503: Determine whether conflict evidence exists according to the conflict coefficient σ.

According to whether the conflict coefficient σ is greater than 0.5, it is judged whether there is conflict evidence, and the conflict degree of the evidence is judged, and the conflict evidence is corrected.

S504: If there is conflicting evidence, modify the basic probability distribution function, and fuse the modified basic probability distribution function through the D-S evidence theory.

If there is conflicting evidence, calculate the weight T _i of the conflicting evidence, and perform weighted average correction on the basic probability distribution function. The method of weighting and modifying the basic probability distribution function is as follows:

The trust degree of evidence is used to describe the trust degree of other evidence to a certain evidence, and it is the degree to which an evidence is supported in the overall evidence. If there are n evidence bodies m ₁ , m ₂ ... m _n , the formula for calculating the trust degree of each evidence is:

The degree of falsity, the degree of falsity is used to describe the degree of falsity of evidence in the whole, the degree of falsity of evidence is relative to the degree of trust, the smaller the degree of trust, the greater the degree of falsity. Under the unified identification framework Θ, their conflicts among multiple evidence sources are:

The conflict of the remaining evidence after removing the jth evidence source from the evidence sources is:

Then the degree of falsity of the evidence is:

According to the degree of trust and falsity, the weight assigned to the focal element of evidence can be obtained as follows:

τ _i =CrdPm _i -γ*Falm _i +1 (15)

In formula (15), if the conflict coefficient σ is greater than 0.5, then γ takes 1 or 2; if the conflict coefficient σ is greater than 0.9, then γ takes

The weight T _i of the conflict evidence after the normalization of the weight distribution of the evidence focal element is:

The basic probability assignment function is modified according to the weight T _i of the conflicting evidence.

After the basic probability distribution function is modified, the D-S combination rule is used to fuse the information of the modified basic probability distribution function to comprehensively diagnose the mechanical failure of the circuit breaker.

S505: If there is no conflict coefficient, directly fuse the basic probability distribution function through the D-S evidence theory.

Using DS evidence theory to fuse the basic probability distribution function, for The Dempster composition rule of two functions m ₁ and m ₂ on U is:

The method for diagnosing mechanical faults of high-voltage circuit breakers based on the D-S evidence theory provided by the embodiment of the present application conducts adaptive neuro-fuzzy control training after performing feature extraction on normal and fault samples of sound signals and vibration signals, and establishes adaptive neuro-fuzzy reasoning The system (ANFIS) model classifies and preprocesses the monitoring data after the feature quantity extraction, and obtains the basic probability assignment function (BPA). The information fusion is carried out through the D-S evidence theory, and the mechanical failure of the circuit breaker is comprehensively diagnosed. Aiming at the conflict problem existing in the fusion process, D-S evidence fusion is carried out after the basic probability assignment function (BPA) is weighted and corrected by the conflict coefficient, which not only does not destroy the credibility of the source evidence, but also improves the convergence speed and Accuracy, fusion results will not appear contrary to the facts. The high-voltage circuit breaker mechanical fault diagnosis method based on the D-S evidence theory provided by this application combines sound and vibration signals to comprehensively diagnose the mechanical fault of the high-voltage circuit breaker, which improves the reliability and accuracy of state assessment.

Other embodiments of the present application will be readily apparent to those skilled in the art from consideration of the specification and practice of the inventive disclosure herein. This application is intended to cover any modification, use or adaptation of the present invention, these modifications, uses or adaptations follow the general principles of the application and include common knowledge or conventional technical means in the technical field not disclosed in the application . The specification and examples are to be considered exemplary only, with the true scope and spirit of the application indicated by the contents of the appended claims.

The embodiments of the present application described above are not intended to limit the scope of protection of the present application.

Claims (7)

1. A high-voltage circuit breaker mechanical fault diagnosis method based on a D-S evidence theory is characterized by comprising the following steps:

establishing a self-adaptive neural fuzzy inference system model;

acquiring a sound signal and a vibration signal of the high-voltage circuit breaker;

extracting characteristic quantities of the sound signal and the vibration signal;

classifying and preprocessing the characteristic quantities respectively through the self-adaptive neural fuzzy inference system model to obtain a basic probability distribution function;

and fusing the basic probability distribution functions through a D-S evidence theory corrected by the high-conflict evidence to obtain a mechanical fault diagnosis result.

2. The method of claim 1, wherein building an adaptive neuro-fuzzy inference system model comprises:

acquiring historical data of sound and vibration signals of the high-voltage circuit breaker;

extracting characteristic quantities of historical data of the sound and vibration signals;

and respectively carrying out self-adaptive neural fuzzy control training on the characteristic quantities and establishing a self-adaptive neural fuzzy inference system model.

3. The method of claim 2, wherein the network structure of the adaptive neuro-fuzzy inference system model comprises a fuzzy layer, a regular inference layer, a normalization layer, an inverse fuzzy layer, and an output layer, wherein,

the fuzzification layer fuzzifies the input characteristic quantity, and the output function of the node i is as follows:

in the formula (1), x is the input of the node i, A_iAnd B_i-2As a fuzzy set, O_1,iMembership function values of a fuzzy set represent that x belongs to A_iSize of degree of (D), μ A_iAnd μ B_i-2Membership functions that are fuzzy sets;

in the formulae (2) and (3), c_i，σ_iIs a membership functionThe parameters of (1);

the rule reasoning layer calculates the excitation degree of each rule, and the output expression is as follows:

in the formula (4), w_iThe excitation strength of the corresponding rule, namely the weight of each fuzzy rule;

the normalization layer calculates the ratio of the excitation intensity of the ith rule to the sum of the excitation intensities of all the rules, and the output expression is as follows:

in the formula (5), the reaction mixture is,the normalized excitation intensity of the ith rule represents the contribution of the ith rule to the final result;

the inverse fuzzy layer calculates the output of each rule, and the output expression is as follows:

in the formula (6), the parameter p_i、q_i、r_iThe back-piece parameters of each node are obtained;

the output layer calculates the sum of all the defuzzified node outputs and outputs,

in the formula (7), Y is the total output of the system, Y_iIs the output of the ith rule.

4. The method according to claim 1, wherein fusing the basic probability distribution functions through a high-collision evidence modified D-S evidence theory to obtain a mechanical fault diagnosis result comprises:

calculating an evidence distance d and a normalization constant k between the evidences;

averaging the evidence distance d and the normalization constant k to obtain a collision coefficient sigma;

judging whether a conflict evidence exists according to the conflict coefficient sigma;

if conflict evidence exists, correcting the basic probability distribution function, and fusing the corrected basic probability distribution function through a D-S evidence theory;

and if no conflict evidence exists, fusing the basic probability distribution functions directly through a D-S evidence theory.

5. The method of claim 4, wherein computing the evidence distance d and the normalization constant k between the evidences comprises:

evidence m₁、m₂The witness distance d between is:

in the formulae (8) and (9), A_i，B_j∈U,m₁(A_i)、m₂(B_j) Respectively, the basic probability assignment of the element A, B, and | A | is the number of the included elements;

the normalized constant k is calculated as:

in the formula (10), k reflects the conflict degree of the evidence, and the value range is [0,1 ].

6. The method of claim 4, wherein revising the basic probability distribution function if conflicting evidence exists comprises:

calculate confidence for each evidence:

in formula (11), sim (m)_i,m_j) As evidence m_i、m_jSimilarity between them, sim (m)_i,m_j)＝1-d(m_i,m_j)；

Under a unified recognition framework, conflicts among multiple evidence sources are computed:

the conflict of the remaining evidence after removing the jth evidence source from the evidence source is:

according to the conflict k₀And conflict k_jCalculating the false degree of the obtained evidence:

calculating according to the trust degree and the false degree to obtain the weight distributed by the evidence focal element as follows:

τ_i＝CrdPm_i-γ*Falm_i+1 (15)

in the formula (15), if the coefficient of collision σ is greater than 0.5, γ is 1 or 2; if the conflict coefficient sigma is larger than 0.9, taking gamma as 2;

normalizing the weight to obtain the weight T of the conflict evidence_i：

According to the weight T_iAnd correcting the basic probability distribution function.

7. The method according to claim 4, wherein the rule for fusing the basic probability distribution functions by D-S evidence theory is:

wherein,