CN101726688A - Method for diagnosing multi-data-source information fusion-based power system fault - Google Patents

Abstract

The invention discloses a method for diagnosing a power system fault, which analyzes various fault information of an electric value, a switching value and the like, performs information fusion on the fault information based on a D-S evidence theory, performs a diagnosis decision on the fault information by a C-mean algorithm, and finally obtains a fault diagnosis result. The method comprises the following steps of: acquiring the fault information (the electric value and the switching value); preprocessing a fault; quantifying electric value information and switching value information by using wavelet analysis technology and fuzzy Petri network technology; extracting fault characteristics from the fault information; adding uncertainty to the characteristics of each type of faults so as to form an evidence body of the D-S evidence theory; performing the information fusion on the evidence body based on the D-S evidence theory; and performing the diagnosis decision by the C-mean algorithm so as to obtain a more precise comprehensive diagnosis result.

Description

A kind of electric network failure diagnosis method of multi-data source information fusion

Technical field

The present invention relates to dispatching of power netwoks and fault analysis field, relate in particular to a kind of electric network failure diagnosis method.

Background technology

Along with the continuous development of the computing machine and the communication technology, the application of intelligent electronic devices such as digital protecting and fault oscillograph in electrical network is more and more general.When electrical network broke down, obtaining of various failure messages was also more convenient.Electric network failure diagnosis method in the past is mostly based on switching value information, expert system, neural network, optimized Algorithm, bayes method and Petri net etc. are arranged, these methods have only been utilized switching value information and have been underused electric parameters information, and electric parameters information has the incomparable advantage of switching value information at aspects such as accuracy, completeness and fault-tolerances.

When electric system is broken down, at first be that electric parameters such as the electric current of fault element and voltage change, cause the protection action then, at last by the corresponding switch of protection tripping.But because protection, switch exist malfunction, tripping in some cases and because of many uncertain factors such as channel disturbance generation information dropouts, and above-mentioned in this case method for diagnosing faults is difficult to obtain correct diagnostic result.Just seem particularly necessary so when electric system is broken down, utilize the failure wave-recording electric parameters to diagnose.Utilize the failure wave-recording electric parameters to carry out fault diagnosis and can avoid interference in the relay protection process, can directly utilize failure message to come analysis of failure.

Summary of the invention

At the deficiencies in the prior art part, the invention provides a kind of method for diagnosing faults of electrical network,

The technical solution adopted in the present invention is: a kind of method for diagnosing faults of electrical network may further comprise the steps:

1, obtains failure message (electric parameters information, switching value information);

2, failure message pre-service: based on wavelet analysis technology and Fuzzy Petri Net technology failure message is analyzed, extracted fault signature, form fault signatures such as wavelet singular degree, small echo fault degree, wavelet energy degree and fuzzy fault degree;

Wavelet singular degree: after fault takes place, fault-signal obtains the wavelet conversion coefficient matrix through behind the wavelet transformation, obtain the singular value features matrix of wavelet coefficient matrix according to the svd Theoretical Calculation, it can represent the basic mode feature of wavelet conversion coefficient matrix.If Λ _i=diag (λ ₁, λ ₂... λ _l) be the i of system (i=1,2 ... n) the singular value features matrix of individual element, order $S_i=\frac{\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\λ}}_i}{l},$ Utilize formula $x_i=\frac{V_i^2}{V_1^2+V_2^2+......V_n^2}$ To S _i(i=1 ... n) do following processing: $m_i=\frac{S_i^2}{S_1^2+S_2^2+......+S_n^2},$ M then _iBe called the wavelet singular degree (WSD, Wavelet Singularity Degree) that i the element in back takes place fault.

Small echo fault degree: when the system of setting up departments breaks down, i (i=1 ... n) fault-signal of individual element acquisition is x _i(n), the wavelet transformation that obtains multiresolution analysis D as a result _I1, D _I2D _Il, wherein l represents that signals sampling counts.D _I1, D _I2D _IkBe the wavelet transformation result of fault front signal correspondence, D _{I (k+1)}, D _{I (k+2)}D _IlWavelet transformation result for signal correspondence after the fault.Order $\left\{\begin{array}{c}{F}_{\mathrm{if}}=\max \{D_{i1},D_{i2}......D_{\mathrm{ik}}\}\\ F_{\mathrm{ib}}=\max \{D_{i(k+1)},D_{i(k+2)}......D_{\mathrm{il}}\}\end{array}\right.,$ $V_i=\frac{\max \{F_{\mathrm{if}},F_{\mathrm{ib}}\}}{\min \{F_{\mathrm{if}},F_{\mathrm{ib}}\}}$ V wherein _iThe intensity of variation of representing the amplitude of signal before and after fault, but it only represents that the amplitude of element fault front and back signal changes, and can not characterize the fault degree of support of element fully comprehensively, simultaneously can not be as the evidence body of evidence theory fusion.Now to V _i(i=1 ... n) carry out following processing $x_i=\frac{V_i^2}{V_1^2+V_2^2+......V_n^2}$ X wherein _iBe called the small echo fault degree (WFD, Wavelet Fault Degree) that i the element in back takes place fault.

Wavelet energy degree: fault-signal x _i(n) carry out the wavelet transformation of multiresolution analysis, establish E ₁, E ₂..., E _mFor the wavelet energy of signal on m yardstick distributes.Wherein $E_j=\underset{k}{\mathrm{\Σ}}{\left|D_j\left(k\right)\right|}^2(j=1......m,k=1......l).$ Order $W=\frac{\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{E}_j}{m},$ $e_i=\frac{W_i^2}{W_1^2+W_2^2+......W_n^2}$ Obtain wavelet energy degree (WED, the Wavelet EnergyDegree) e of signal _iCome the strong and weak degree of characterization signal energy.

Fuzzy fault degree: according to the switching value information behind the electric network fault (protection, isolating switch), utilize the Fuzzy Petri Net technology that electrical network is diagnosed, obtain the fault signature P of each element _i(i=1 ... n), order $y_i=\frac{P_i^2}{P_1^2+P_2^2+......P_n^2},$ Y wherein _iBe called the fuzzy fault degree (FFD, FuzzyFault Degree) that i the element in back takes place fault.

3, add uncertainty for each fault signature, form the evidence body;

If F is the identification framework of electric network failure diagnosis, and F comprises n element, wherein F _iThe malfunction of representing i element, then m is the basic reliability distribution on the Fault Identification framework F, m (F _i) be called F _iCredible substantially number.M (F _i) characterize i the probabilistic tolerance of element fault state.

In invention, as evidence body structure basic reliability distribution function independently, their expressions are represented with x the degree of support of element fault with small echo fault degree, wavelet singular degree, wavelet energy degree and fuzzy fault degree.

If the Fault Identification framework comprises q bar evidence, the number of element to be identified is n in the identification framework, then $m_j\left(F_i\right)=\frac{(1-u_j)x_{\mathrm{ij}}}{x_j}$ Wherein, i=1 ... n; J=1 ... q; x _IjIt is the fault degree of support of the j class evidence body of i element correspondence.x _jBe the fault support sum of each element of j class evidence body, u _jBe the uncertainty of j class evidence body, wherein the uncertainty of wavelet singular degree, small echo fault degree and wavelet energy degree is 0.1, and the uncertainty of fuzzy fault degree is 0.15.m _j(F _i) be the basic confidence level of the j class evidence body correspondence of i element fault.

4, based on the D-S evidence theory each evidence body is carried out information fusion, obtaining fusion results is the probability of malfunction sign of each element;

5, based on the C-averaging method fusion results is diagnosed decision-making, obtain final diagnostic result.

1) probability of malfunction of n element characterizes and is respectively m (F ₁), m (F ₂) ... m (F _n).All elements initially are divided into two classes, get the Gamma function $\mathrm{\&Gamma;}\left(x\right)={\&Integral;}_0^{\&infin;}{t}^{x-1}{e}^{-t}\mathrm{dt},$ Each element fault probability tables solicited get the Gamma functional value, if satisfy condition: $\mathrm{\&Gamma;}\left(m\left(F_i\right)\right)={\&Integral;}_0^{\&infin;}{t}^{m\left(F_i\right)-1}{e}^{-t}\mathrm{dt}<\mathrm{\&epsiv;},$ Be classified as fault element candidate class Γ ₁The residue element is classified as non-fault element candidate class Γ ₂, ε=12 wherein;

2) in the fault candidate class N is arranged ₁Individual element, its corresponding Gamma functional value is respectively

Get $E=\min \{E_1,E_2......E_{N_1}\}$ Pairing element is absolute failure element (necessarily breaking down);

3) to the preliminary classification (Γ of all elements ₁, Γ ₂), calculate its average, last error of calculation quadratic sum sorting criterion; $\left\{\begin{array}{c}{m}_i=\frac{1}{N_i}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}m\left(F_j\right)\\ (i=\mathrm{1,2};j=1......N_i)\end{array}\right.,$ $J_e=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2,$ J _eMeasured the total square-error that is produced when representing its category set, made J with the classification average _eMinimum classification is a classification results optimum under the error sum of squares criterion.

4) from Γ _iThe middle sample m (F that selects _j);

5) if N _i=1, then change step 4), otherwise continue;

6) calculate $\left\{\begin{array}{c}{\mathrm{\&rho;}}_k=\frac{N_k}{N_k+1}{\left|\right|m\left(F_j\right)-m_k\left|\right|}^2\\ {\mathrm{\&rho;}}_i=\frac{N_i}{N_i-1}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2\end{array}\right.$

7) if satisfy ρ _k≤ ρ _i, then m (F _j) from Γ _iMove on to Γ _kIn go; Recomputate m _iAnd m _kValue, and revise J _e

8) if N J of subsequent iteration _eDo not change, then stop, otherwise the element that forwards in the step 4. fault candidate class is fault element.

The present invention proposes the electric network failure diagnosis system of multi-data source information fusion, analyze various faults information such as electric parameters, switching value, carry out information fusion, adopt the C-mean algorithm to diagnose decision-making, finally obtain fault diagnosis result based on the D-S evidence theory.

The invention has the beneficial effects as follows: the present invention combines existing data acquisition equipment, proven technique with the knowwhy in forward position, the electric network failure diagnosis system of multi-data source information fusion has been proposed, analyze various faults information such as electric parameters, switching value, carry out information fusion based on the D-S evidence theory, adopt the C-mean algorithm to diagnose decision-making, finally obtain fault diagnosis result.This method is based on the analysis in switch amount and electric parameters Double Data source, efficiently solve defectives such as the information that switching value information forms data source faces is inaccurate, information is incomplete, by Fuzzy Petri Net and wavelet analysis failure message is carried out feature extraction, obtain corresponding fuzzy fault degree and small echo fault signature (wavelet singular degree, small echo fault degree, wavelet energy degree).They can accurately characterize fault-signal, and carry out information fusion on D-S evidence theory basis, adopt the C-mean algorithm to diagnose decision-making, and then obtain accurate more comprehensive diagnos result.

Embodiment

Now the present invention is described further, it is more obvious that purpose of the present invention and effect will become:

The method for diagnosing faults of electrical network of the present invention may further comprise the steps:

1, obtains failure message (electric parameters information, switching value information);

2, failure message pre-service: based on wavelet analysis technology and Fuzzy Petri Net technology failure message is analyzed, extracted fault signature, form fault signatures such as wavelet singular degree, small echo fault degree, wavelet energy degree and fuzzy fault degree;

Wavelet singular degree: after fault takes place, fault-signal obtains the wavelet conversion coefficient matrix through behind the wavelet transformation, obtain the singular value features matrix of wavelet coefficient matrix according to the svd Theoretical Calculation, it can represent the basic mode feature of wavelet conversion coefficient matrix.If Λ _i=diag (λ ₁, λ ₂... λ _n) be the i of system (i=1,2 ... n) the singular value features matrix of individual element, order $S_i=\frac{\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i}{l},$ Utilize formula $x_i=\frac{V_i^2}{V_1^2+V_2^2+......V_n^2}$ To S _i(i=1 ... n) do following processing: $m_i=\frac{S_i^2}{S_1^2+S_2^2+......S_n^2},$ M then _iBe called the wavelet singular degree (WSD, Wavelet Singularity Degree) that i the element in back takes place fault.

Small echo fault degree: when the system of setting up departments breaks down, i (i=1 ... n) fault-signal of individual element acquisition is x _i(n), the wavelet transformation that obtains multiresolution analysis D as a result _I1, D _I2D _Il, wherein l represents that signals sampling counts.D _I1, D _I2D _IkBe the wavelet transformation result of fault front signal correspondence, D _{I (k+1)}, D _{I (k+2)}D _IlWavelet transformation result for signal correspondence after the fault.Order $\left\{\begin{array}{c}{F}_{\mathrm{if}}=\max \{D_{i1},D_{i2}......D_{\mathrm{ik}}\}\\ F_{\mathrm{ib}}=\max \{D_{i(k+1)},D_{i(k+2)}......D_{\mathrm{il}}\}\end{array}\right.,$ $V_i=\frac{\max \{F_{\mathrm{if}},F_{\mathrm{ib}}\}}{\min \{F_{\mathrm{if}},F_{\mathrm{ib}}\}}$ V wherein _iThe intensity of variation of representing the amplitude of signal before and after fault, but it only represents that the amplitude of element fault front and back signal changes, and can not characterize the fault degree of support of element fully comprehensively, simultaneously can not be as the evidence body of evidence theory fusion.Now to V _i(i=1 ... n) carry out following processing $x_i=\frac{V_i^2}{V_1^2+V_2^2+......V_n^2}$ X wherein _iBe called the small echo fault degree (WFD, Wavelet Fault Degree) that i the element in back takes place fault.

Wavelet energy degree: fault-signal x _i(n) carry out the wavelet transformation of multiresolution analysis, establish E ₁, E ₂..., E _mFor the wavelet energy of signal on m yardstick distributes.Wherein $E_j=\underset{k}{\mathrm{\&Sigma;}}{\left|D_j\left(k\right)\right|}^2(j=1......m,k=1......l).$ Order $W=\frac{\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{E}_j}{m},$ $e_i=\frac{W_i^2}{W_1^2+W_2^2+......W_n^2}$ Obtain wavelet energy degree (WED, the Wavelet EnergyDegree) e of signal _iCome the strong and weak degree of characterization signal energy.

Fuzzy fault degree: according to the switching value information behind the electric network fault (protection, isolating switch), utilize the Fuzzy Petri Net technology that electrical network is diagnosed, obtain the fault signature P of each element _i(i=1 ... n), order $y_i=\frac{P_i^2}{P_1^2+P_2^2+......P_n^2},$ Y wherein _iBe called the fuzzy fault degree (FFD, FuzzyFault Degree) that i the element in back takes place fault.

3, add uncertainty for each fault signature, form the evidence body;

If F is the identification framework of electric network failure diagnosis, and F comprises n element, wherein F _iThe malfunction of representing i element, then m is the basic reliability distribution on the Fault Identification framework F, m (F _i) be called F _iCredible substantially number.M (F _i) characterize i the probabilistic tolerance of element fault state.

In invention, as evidence body structure basic reliability distribution function independently, their expressions are represented with x the degree of support of element fault with small echo fault degree, wavelet singular degree, wavelet energy degree and fuzzy fault degree.

If the Fault Identification framework comprises q bar evidence, the number of element to be identified is n in the identification framework, then $m_j\left(F_i\right)=\frac{(1-u_j)x_{\mathrm{ij}}}{x_j}$ Wherein, i=1 ... n; J=1 ... q; x _IjIt is the fault degree of support of the j class evidence body of i element correspondence.x _jBe the fault support sum of each element of j class evidence body, u _jBe the uncertainty of j class evidence body, wherein the uncertainty of wavelet singular degree, small echo fault degree and wavelet energy degree is 0.1, and the uncertainty of fuzzy fault degree is 0.15.m _j(F _i) be the basic confidence level of the j class evidence body correspondence of i element fault.

4, based on the D-S evidence theory each evidence body is carried out information fusion, obtaining fusion results is the probability of malfunction sign of each element;

5, based on the C-averaging method fusion results is diagnosed decision-making, obtain final diagnostic result.

1) probability of malfunction of n element characterizes and is respectively m (F ₁), m (F ₂) ... m (F _n).All elements initially are divided into two classes, get the Gamma function $\mathrm{\&Gamma;}\left(x\right)={\&Integral;}_0^{\&infin;}{t}^{x-1}{e}^{-t}\mathrm{dt},$ Each element fault probability tables solicited get the Gamma functional value, if satisfy condition: $\mathrm{\&Gamma;}\left(m\left(F_i\right)\right)={\&Integral;}_0^{\&infin;}{t}^{m\left(F_i\right)-1}{e}^{-t}\mathrm{dt}<\mathrm{\&epsiv;},$ Be classified as fault element candidate class Γ ₁The residue element is classified as non-fault element candidate class Γ ₂, ε=12 wherein;

2) in the fault candidate class N is arranged ₁Individual element, its corresponding Gamma functional value is respectively

Get $E=\min \{E_1,E_2......E_{N_1}\}$ Pairing element is absolute failure element (necessarily breaking down);

3) to the preliminary classification (Γ of all elements ₁, Γ ₂), calculate its average, last error of calculation quadratic sum sorting criterion; $\left\{\begin{array}{c}{m}_i=\frac{1}{N_i}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}m\left(F_j\right)\\ (i=\mathrm{1,2};j=1......N_i)\end{array}\right.,$ $J_e=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2,$ J _eMeasured the total square-error that is produced when representing its category set, made J with the classification average _eMinimum classification is a classification results optimum under the error sum of squares criterion.

4) from Γ _iThe middle sample m (F that selects _j);

5) if N _i=1, then change step 4), otherwise continue;

6) calculate $\left\{\begin{array}{c}{\mathrm{\&rho;}}_k=\frac{N_k}{N_k+1}{\left|\right|m\left(F_j\right)-m_k\left|\right|}^2\\ {\mathrm{\&rho;}}_i=\frac{N_i}{N_i-1}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2\end{array}\right.$

7) if satisfy ρ _k≤ ρ _i, then m (F _j) from Γ _iMove on to Γ _kIn go; Recomputate m _iAnd m _kValue, and revise J _e

8) if N J of subsequent iteration _eDo not change, then stop, otherwise the element that forwards in the step 4. fault candidate class is fault element.

The present invention proposes the electric network failure diagnosis system of multi-data source information fusion, analyze various faults information such as electric parameters, switching value, carry out information fusion, adopt the C-mean algorithm to diagnose decision-making, finally obtain fault diagnosis result based on the D-S evidence theory.The diagnostic system framework as shown in Figure 1.

The invention provides a kind of method for diagnosing faults of electrical network, existing data acquisition equipment, proven technique are combined with the knowwhy in forward position, the electric network failure diagnosis system of multi-data source information fusion has been proposed, analyze various faults information such as electric parameters, switching value, carry out information fusion based on the D-S evidence theory, adopt the C-mean algorithm to diagnose decision-making, finally obtain fault diagnosis result.This method is based on the analysis in switch amount and electric parameters Double Data source, efficiently solve defectives such as the information that switching value information forms data source faces is inaccurate, information is incomplete, by Fuzzy Petri Net and wavelet analysis failure message is carried out feature extraction, obtain corresponding fuzzy fault degree and small echo fault signature (wavelet singular degree, small echo fault degree, wavelet energy degree).They can accurately characterize fault-signal, and carry out information fusion on D-S evidence theory basis, adopt the C-mean algorithm to diagnose decision-making, and then obtain accurate more comprehensive diagnos result.

Claims (2)

1. the method for diagnosing faults of an electrical network is characterized in that, may further comprise the steps:

(1) obtains failure message.

(2) failure message pre-service: based on wavelet analysis technology and Fuzzy Petri Net technology failure message is analyzed, extracted fault signature, form fault signatures such as wavelet singular degree, small echo fault degree, wavelet energy degree and fuzzy fault degree.

(3) add uncertainty for each fault signature, form the evidence body.

(4) based on the D-S evidence theory each evidence body is carried out information fusion, obtaining fusion results is the probability of malfunction sign of each element.

(5) based on the C-averaging method fusion results is diagnosed decision-making, obtain final diagnostic result.

2. according to the method for diagnosing faults of the described electrical network of claim 1, it is characterized in that described step (5) is specially:

(A) probability of malfunction of n element characterizes and is respectively m (F ₁), m (F ₂) ... m (F _n); All elements initially are divided into two classes, get the Gamma function $\mathrm{\&Gamma;}\left(x\right)={\&Integral;}_0^{\&infin;}{t}^{x-1}{e}^{-t}\mathrm{dt},$ Each element fault probability tables solicited get the Gamma functional value, if satisfy condition: $\mathrm{\&Gamma;}\left(m\left(F_i\right)\right)={\&Integral;}_0^{\&infin;}{t}^{m\left(F_i\right)-1}{e}^{-t}\mathrm{dt}<\mathrm{\&epsiv;}$ Be classified as fault element candidate class Γ ₁The residue element is classified as non-fault element candidate class Γ ₂, ε=12 wherein.

(B) in the fault candidate class N is arranged ₁Individual element, its corresponding Gamma functional value is respectively

Get $E=\min \{E_1,E_2\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;E_{N_1}\}$ Pairing element is the absolute failure element.

(C) to the preliminary classification (Γ of all elements ₁, Γ ₂), calculate its average, last error of calculation quadratic sum sorting criterion; $\left\{\begin{array}{c}{m}_i=\frac{1}{N_i}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}m\left(F_j\right)\\ (i=\mathrm{1,2};j=1\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;N_i)\end{array}\right.,$ $J_e=\underset{i=1}{\overset{2}{\mathrm{\&Sigma;}}}\underset{m\left(F_j\right)\&Element;{\mathrm{\&Gamma;}}_i}{\mathrm{\&Sigma;}}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2,$ J _eMeasured the total square-error that is produced when representing its category set, made J with the classification average _eMinimum classification is a classification results optimum under the error sum of squares criterion.

(D) from Γ _iThe middle sample m (F that selects _j).

(E) if N _i=1, then change step 4), otherwise continue.

(F) calculate $\left\{\begin{array}{c}{\mathrm{\&rho;}}_k=\frac{N_k}{N_k+1}{\left|\right|m\left(F_j\right)-m_k\left|\right|}^2\\ {\mathrm{\&rho;}}_i=\frac{N_i}{N_i-1}{\left|\right|m\left(F_j\right)-m_i\left|\right|}^2\end{array}\right..$

(G) if satisfy ρ _k≤ ρ _i, then m (F _j) from Γ _iMove on to Γ _kIn go; Recomputate m _iAnd m _kValue, and revise J _e

(H) if N J of subsequent iteration _eDo not change, then stop, otherwise the element that forwards in the step 4. fault candidate class is fault element.