CN101576604B - Method for diagnosing failures of analog circuit based on heterogeneous information fusion - Google Patents

Abstract

The invention discloses a method for diagnosing failures of an analog circuit based on heterogeneous information fusion. The method comprises the following steps: measuring a temperature variation value of a circuit element to be diagnosed; calculating an optimum gate valve value of the variation value of working temperature of each element; integrating the temperature variation value of an element more than the optimum gate valve value into a failure domain phi, and calculating by a temperature membership grade model to acquire a temperature evidence value of each element; applying an actuating signal to a circuit to be diagnosed to measure voltage of a measurable point; inputting information of the voltage of the measurable point into a BP neural network to carry out primary diagnosis soas to acquire an evidence value of voltage of the measurable point; and fusing the evidence value of temperature and the evidence value of voltage by a heterogeneous information fusion system, acquir ing an associated weight coefficient, identifying the weight coefficient and unifying the weight coefficient, and regulating and carrying out D-S fusion on the two types of the evidence values to determine a failure element. The method solves the problem of inaccurate judgment caused by deficiency of feature information and incompatible heterogeneous information in failure diagnosis of the analog circuit, and improves the accuracy of the failure diagnosis.

Description

Analog-circuit fault diagnosis method based on the heterogeneous information fusion

Technical field

The present invention relates to a kind of mimic channel diagnostic method, particularly a kind of analog-circuit fault diagnosis method that merges based on heterogeneous information.

Background technology

Analog circuit fault diagnosing research is through the development of four more than ten years; Made significant headway; But the continuous reinforcement with integrated level that improves constantly along with the complicated circuit degree; Be difficult to find in the actual test in the reached node, particularly large-scale complex circuit of enough numbers and can survey node still less, the topological structure of side circuit is very undesirable; But the topological structure of circuit and measuring point etc. have restricted the applicability of circuit diagnostics method, and the extensive non-linear and tool tolerance of circuit component and the diversity of phenomenon of the failure thereof etc. make existing diagnostic method be difficult to solve the diagnosis problem of side circuit.

The many difficult problems that solve in the analog circuit fault diagnosing that are applied as of information fusion technology provide possibility; But this research also is in the starting stage; Obtain breakthrough progress and finally form ripe application technology, also possibly need more innovative research or a large amount of careful work of improving.An elementary tactics of evidence blending theory is divided into a plurality of incoherent parts with the evidence set exactly, and utilizes them respectively framework of identification to be judged, merges it with the D-S rule then.Briefly be exactly the judgement that obtains from different aforementioned sources or sensor through each group, merge through evidence and make a strategic decision out near truth to same object event, and the most rational judged result.Though this theory has obtained to use widely; Particularly in pattern-recognition; Important effect has been brought into play in aspects such as Target Recognition; But using it carries out fault diagnosis and still exist problems: 1. because the method robustness is lower, the subtle change of often single evidence body value will cause that combined result produces sharply and change; 2. often disagree for the fusion of conflicting evidence value with the intuition and the fact, obviously unkind and irrigational; 3. complete when inconsistent when an evidence and many evidences, veto by one vote appears after the combination.Special diagnosis for the large-scale circuit system, since sensor, the influence of the complexity of circuit own and other factors, and evidence body itself is just unrealistic accurately to obtain ideal.

Summary of the invention

In order to solve the above-mentioned technical matters that existing analog circuit fault diagnosing exists, the present invention provides a kind of accuracy the high analog-circuit fault diagnosis method based on the heterogeneous information fusion.

The technical method that the present invention solves the problems of the technologies described above may further comprise the steps:

A kind of analog-circuit fault diagnosis method that merges based on heterogeneous information may further comprise the steps:

1) utilize temperature sensor measurement to treat the temperature change value of diagnostic circuit element;

2) ask for the optimum gate threshold values of each element temperature change value, its step is following:

A1) the component temperature changing value information of obtaining according to sensor utilizes iterative formula to calculate family of power and influence's value;

A2) family of power and influence's value that utilize to obtain is cut apart and is handled each element temperature change value; Through the temperature change value information calculations optimized discrete degree that obtains after the dividing processing and calculate element number,, the element number principle that satisfies defined and optimized discrete stop calculating when spending;

A3) then do not carry out next step iterative computation if do not meet the demands, repeat above-mentioned steps a2) and a3), till obtaining the best family of power and influence's value that satisfies condition, thereby obtain the optimum gate threshold values;

3) the optimum gate threshold values filtering element temperature change value that utilize to obtain, when certain element temperature change value then is included into it among failure domain Φ during greater than the optimum gate threshold values, otherwise this temperature variation value information of filtering;

4) temperature change value with failure domain Φ is input in the temperature degree of membership model, utilizes membership function calculate to obtain the fault degree of membership value of each element temperature change value, with it as temperature evidence value;

5) treat diagnostic circuit and apply pumping signal, but measure measuring point voltage;

6) but measuring point voltage input BP neural network is carried out elementary diagnosis, but obtain the evidence value of measuring point voltage, and with it as voltage evidence value;

7) utilize the heterogeneous information emerging system that said temperature evidence value and voltage evidence value are merged, confirm fault element, the steps include:

The evidence collision vector of a, accounting temperature evidence value and voltage evidence value with the normalization of evidence collision vector and ask entropy, is asked the associated weights coefficient that obtains reciprocal again;

B, temperature evidence value, voltage evidence value are compared with desirable fault output vector respectively, calculate both likelihood similarities;

C, according to each group likelihood similarity utilize genetic algorithm to seek to satisfy similarity maximum respectively organize the priori weight coefficient;

D, go out unified weight allocation coefficient according to priori weight coefficient and associated weights coefficient calculations, unified weight coefficient computing formula does

${\mathrm{\ξ}}_{\mathrm{\κj}}=\frac{w_{\mathrm{\κ}}+r_{\mathrm{\κj}}}{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{r}_{\mathrm{ij}}+\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w}_i},j=\mathrm{1,2},\·\·\·,x$

$\mathrm{st}.\underset{\mathrm{\κ}=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ξ}}_{\mathrm{\κj}}=1$

The number of priori weight coefficient, associated weights coefficient and unified weight coefficient is x, ξ in the formula _{κ j}Be illustrated in the j class fault that adds up in the m class and be the unified weight allocation coefficient under the κ class at sensor, but accounting temperature evidence primary system one weight coefficient during with measuring point voltage evidence primary system one weight coefficient k get 1 and 2 respectively, w _κFor being the associated weights coefficient under the κ class at sensor, r _KjFor being that the κ class is the priori weight coefficient in j class fault and at sensor;

E, the unified weight allocation coefficient of utilization are readjusted each evidence value, carry out D-S again and merge the fault location element.

Technique effect of the present invention is: 1) improving constantly along with complicated circuit degree and integrated level; Be difficult to find the reached node of enough numbers in the actual test; Particularly can survey node still less in the large-scale complex circuit, and the topological structure of side circuit is very undesirable, but will extracts very difficulty of sufficient measuring point information of voltage; And present temperature probe has quite high precision; And measuring this kind parameter is contactless on-line measurement mode, does not receive circuit topological structure and can survey the restriction of node, obtains incomplete, the inaccurate inaccurate even wrong problem of localization of fault that causes thereby solved failure message; Improved the accuracy of fault diagnosis; 2) the present invention can extract the temperature variation value information of fault element effectively, has proposed a kind of efficient temperature information extracting method; 3) the present invention combines the temperature and the electric weight failure message of mimic channel; Utilize the heterogeneous information fusion method; Proposed practicable blending algorithm, the misjudgement problem that fusion diagnosis occurs when having solved information collision provides a kind of accuracy high analog-circuit fault diagnosis method.

Below in conjunction with accompanying drawing the present invention is further described.

Description of drawings

Fig. 1 is a process flow diagram of the present invention.

Fig. 2 merges for evidence among the present invention and calculating weight coefficient process flow diagram.

The process flow diagram of Fig. 3 for finding the solution priori weight coefficient method among the present invention.

Fig. 4 calculates the priori weight coefficient for utilizing genetic algorithm among the present invention.

Fig. 5 is the enforcement figure circuit diagram among the present invention.

Embodiment

Referring to Fig. 1, Fig. 1 is a process flow diagram of the present invention.Concrete two category informations that extract of the inventive method can measuring point information of voltage and component temperature changing value information.Total thinking that temperature fault changing value information is obtained: the element that changes according to a plurality of working temperatures of available circuit; Obtain its temperature change value Δ t; The temperature door threshold values ε that component temperature changing value Δ t and is set compares; Element temperature change value then is included among the failure domain Φ during greater than the optimum gate threshold values; The temperature change value of failure domain Φ is input in the temperature degree of membership model, utilizes membership function calculate to obtain the fault degree of membership value of each element temperature change value, with it as temperature evidence value.

In order to extract the efficient temperature failure message, obtain temperature evidence value, family of power and influence's value need be set so that effective filtration temperature fault changing value information.In theory; Though temperature fault changing value information is many more in the set; Fault element will be encompassed in wherein more easily, but also can cause localization of fault more difficulty, situation such as syncretizing effect is undesirable, therefore the rational family of power and influence is set be worth and make and can farthest filter unnecessary component temperature changing value information; Obtain simplifying back component temperature changing value ensemble of communication, and the temperature change value information of its fault element is comprised wherein just highly significant.

The calculation procedure that the family of power and influence is worth ε is following:

If the ensemble of communication of fault element temperature change value is Θ,<img file="GDA00001818625000051.GIF" he="64" img-content="drawing" img-format="GIF" inline="yes" orientation="portrait" wi="552" />And 0<Δ t<sub >1</sub><Δ t<sub >2</sub><<Δ t<sub >n</sub>,, family of power and influence's value obtains set after cutting apart filtration for failure domain Φ, then have

<math> <mrow> <mo>&amp;Exists;</mo> <mo>{</mo> <mi>&amp;Delta;t</mi> <mo>|</mo> <mi>&amp;Delta;</mi> <msub> <mi>t</mi> <mn>1</mn> </msub> <mo>,</mo> <mi>&amp;Delta;</mi> <msub> <mi>t</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>&amp;CenterDot;</mo> <mo>&amp;CenterDot;</mo> <mo>&amp;CenterDot;</mo> <mo>,</mo> <mi>&amp;Delta;</mi> <msub> <mi>t</mi> <mi>m</mi> </msub> <mo>}</mo> <mo>=</mo> <mi>&amp;Phi;</mi> <mrow> <mo>(</mo> <mi>m</mi> <mo>&lt;;</mo> <mi>n</mi> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow></math>

ε wherein<Δ t ₁<Δ t ₂<<Δ t _m, card (Φ) ∈ [1,3], the number of element among card (Φ) the expression failure domain Φ, n is original temperature variation value information number, m can know that by following formula the m number should be not more than 3 for be worth the temperature change value number after iteration is screened through the family of power and influence.Utilize iterative formula to ask the family of power and influence to be worth ε below; If card (Θ)=n; Card (Φ)=m, iterative formula is following:

ε ₁=|T/card(Θ)-max{ΔtΔt ₁,Δt ₂,…,Δt _m}|+T/ccard(Θ)

${\mathrm{\&epsiv;}}_2=\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{\&Delta;}{t}_m\}-{\mathrm{\&epsiv;}}_1|}{2}+{\mathrm{\&epsiv;}}_1---\left(2\right)$

${\mathrm{\&epsiv;}}_h=\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{\&Delta;}{t}_m\}-{\mathrm{\&epsiv;}}_{h-1}|}{2}+{\mathrm{\&epsiv;}}_{h-1}$

Wherein, ε ₁If family of power and influence's numerical value after the expression iteration first time is ε ₁Do not satisfy judgment condition 1. with 2., then continue iteration, h is a number of iterations, when satisfying condition in h step 1. with 2. the time, then termination of iterations, the ε of this moment so _hBe best family of power and influence's value just.Judgment condition is made up of two parts:

1.. when the ε that obtains can make that element number is not more than 3 in failure domain Φ, and when the arbitrary temp changing value all is not less than the family of power and influence and is worth among the failure domain Φ, termination of iterations.

2.. cut apart the failure domain Φ that obtains by ε and should satisfy following formula: promptly the ratio of element number is less than L among the difference of maximum temperature changing value and minimum temperature changing value and the failure domain Φ, and wherein L gets 30-50.

$\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{\&Delta;}{t}_m\}-\min \left\{\mathrm{\&Delta;t}\right|\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,\mathrm{\&Delta;}{\mathrm{<figure-callout\; id="2"\; label="t"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">t</figure-callout>}}_2,\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{\&Delta;}{t}_m\left\}\right|}{\mathrm{card}\left(\mathrm{\&Phi;}\right)}\&le;L---\left(3\right)$

When through after utilizing the optimum gate threshold segmentation to filter to obtain temperature fault changing value information set Φ, wherein each element is input to down formula, the acquisition degree of membership.Confirm that the membership function model is:

<math> <mrow> <msub> <mi>F</mi> <mi>ij</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mfrac> <mrow> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mrow> </mfrac> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <msup> <mrow> <mo>(</mo> <mfrac> <msub> <mi>x</mi> <mi>j</mi> </msub> <mi>&amp;lambda;&amp;gamma;</mi> </mfrac> <mo>)</mo> </mrow> <mi>&amp;gamma;</mi> </msup> <mi>exp</mi> <mrow> <mo>(</mo> <mi>&amp;gamma;</mi> <mo>-</mo> <mfrac> <msub> <mi>x</mi> <mi>j</mi> </msub> <mi>&amp;lambda;</mi> </mfrac> <mo>)</mo> </mrow> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mo>-</mo> <mfrac> <mrow> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mrow> </mfrac> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>></mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow></math>

Wherein, x ₀Be the circuit working normal parameter value of detected element just often; e _IjFor waiting to diagnose the normal variation scope of component parameters; t _IjFor waiting to diagnose the limit deviation of component parameters; F _IjFor sensor j measures the fault degree of membership that measurand belongs to the i pattern; x _jCircuit component temperature change value and x for j class sensor test _j∈ Φ; And λ γ=max (Δ t ₁, Δ t ₂..., Δ t _m).

Through above-mentioned steps, we have obtained the fault degree of membership value about element temperature change value, i.e. temperature evidence value.But we adopt common method promptly to obtain the measuring point information of voltage fault of circuit through the node voltage the surveyed signal that adopts the excitation of multi-frequency AC signal to record down then, are entered into BP again and calculate neural network, but obtain measuring point information of voltage fault evidence value.Wherein but the number of BP neural network input layer is exactly the number of circuit measuring point information of voltage; Output layer is the number of analog circuit element; Confirming of hidden layer number usually according to experimental formula, i.e. ni=sqrt (n0+n1)+a formula, in the formula: ni is implicit node number; N0 is an input number of nodes; N1 is the output node number; A gets constant between 1～10.

But the fault degree of membership value that we have obtained element temperature change value through above-mentioned steps is temperature evidence value and measuring point information of voltage fault evidence value, and we are entered into and carry out Decision Fusion in the heterogeneous information emerging system to obtain final localization of fault.Its particular flow sheet is seen accompanying drawing 2.Carry out information fusion, at first need calculate priori weight coefficient and associated weights coefficient, its concrete steps are following:

At first, we briefly introduce the definition and the physical meaning of priori weight coefficient and associated weights coefficient.According to priori the evidence from different aforementioned sources is provided a support A _s, different like this evidences has different supports.Support on evidence can constitute a support vector, this support vector constitutes so

R′ _f=(A ₁,A ₂,…,A _k) A _s∈(0,1) ?（5）

In the formula, R ' _fFor absolute priori weight, absolute priori weight is done normalization, have:

$w_c=A_k/\underset{k=1}{\overset{s}{\mathrm{\&Sigma;}}}{A}_k---\left(6\right)$

Can obtain a new vector

R _f=(r ₁,r ₂,…,r _k) （7）

In the formula, r _lBe the list value of relative priori weight, and r _l∈ (0; 1). the relative priori weight that so obtains is called the priori weight coefficient at this. introduce the associated weights coefficient according to " majority rule "; In the evidence building-up process; Cause serious conflict or the influence of certain or minority evidence of conflict is less fully, so its weight coefficient is just little.Produce evidence at the same time in This document assumes that x different card source, its evidence collection is E={E ₁, E ₂..., E _x, the associated weights partition factor that then need confirm does

W _f=(w ₁,w ₂,…,w _x) （8）

Satisfy w in the formula _o∈ [0,1] and

The significance level that the evidence that associated weights coefficient reflection evidence source provides distributes according to conflict spectrum each other in building-up process and they are to synthetic result's influence degree.

The method for solving of 1 associated weights coefficient:

According to above-mentioned; But we have obtained temperature evidence value and measuring point information of voltage fault evidence value; We will merge these two types of evidence value sets, but therefore find the solution the associated weights coefficient about temperature evidence value and measuring point information of voltage fault evidence value, at first accounting temperature evidence value E earlier _dBut with measuring point information of voltage fault evidence value E _e(e=1,2 ..., d-1, d+1 ..., the conflict spectrum between x)

Can constitute collision vector

${\&PartialD;}_d=({\&PartialD;}_{\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}1},{\&PartialD;}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}2},\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dd}-1},{\&PartialD;}_{\mathrm{dd}+1},\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dx}})---\left(9\right)$

Wherein

${\&PartialD;}_{\mathrm{De}}=\underset{A_d\&Element;E_d,A_e\&Element;E_e}{\underset{A_d\&cap;A_e=\mathrm{\&phi;}}{\mathrm{\&Sigma;}}}{m}_d\left(A_d\right)m_e\left(A_e\right)$ And d=1,2 ..., x (10)

The temperature evidence value E that collision vector can be described _dBut with measuring point information of voltage fault evidence value E _eBetween the magnitude relationship of conflict, normalization is carried out in (10) gets:

${\&PartialD;}_d^N=\frac{({\&PartialD;}_{\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}1},{\&PartialD;}_{\mathrm{<figure-callout\; id="2"\; label="d"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}2},\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dd}-1},{\&PartialD;}_{\mathrm{dd}+1},\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dx}})}{\underset{e=1,d\&NotEqual;e}{\overset{x}{\mathrm{\&Sigma;}}}{\&PartialD;}_{\mathrm{de}}}=({\&PartialD;}_{d1}^N,{\&PartialD;}_{d2}^N,\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dd}-1}^N,{\&PartialD;}_{\mathrm{dd}+1}^N,\&CenterDot;\&CenterDot;\&CenterDot;,{\&PartialD;}_{\mathrm{dx}}^N)---\left(11\right)$

Each

calculated its entropy

$H_d=\underset{e=1,d\&NotEqual;e}{\overset{x}{\mathrm{\&Sigma;}}}{\&PartialD;}_{\mathrm{de}}^N\ln \left({\&PartialD;}_{\mathrm{de}}^N\right)(d=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,x)---\left(12\right)$

Inverse is got in (12)

$H_d^{-1}=\frac{1}{H_d}---\left(13\right)$

Then but temperature evidence value with the associated weights coefficient of measuring point information of voltage fault evidence value is:

$w_d=\frac{H_d^{-1}}{\underset{e=1}{\overset{x}{\mathrm{\&Sigma;}}}{H}_e^{-1}}---\left(14\right)$

The method for solving of 2 priori weight coefficients:

Generally, because the complicacy of Circuits System, but comparatively sensitive for some fault measuring point information of voltage reflection; Some then temperature information reflect responsively more, therefore for different faults, the information that each sensor obtains is often different; Its accuracy and confidence that provides the result also there are differences, so before before the sensor test data, promptly obtaining an evidence, can be by historical data; Means such as expertise provide a weight, and this weight just is called the priori weight.But accompanying drawing 3 has provided the process flow diagram of the priori weight coefficient of asking for temperature evidence value and measuring point information of voltage fault evidence value.

Same, but because above-mentioned temperature evidence value and the measuring point information of voltage fault evidence value of having obtained, we put into same evidence body E with above-mentioned two types of evidence values _iIn, E wherein _I1, E _I2..., E _IxFor for the evidence that obtains on the i class fault, E ⁱBe desirable evidential probability, and make D-S blending algorithm operator do about i class fault For merging with the D-S amalgamation mode.But we ask for temperature evidence value and measuring point information of voltage fault evidence value priori weight coefficient, but then should exist for the priori weight coefficient R of i class fault about temperature evidence value and measuring point information of voltage fault evidence value _If=(r _I1, r _I2..., r _Ik), make:

Max：F=Sup(E ⁱ,E ^* _i) （15）

$s.t{E^{*}}_i=\stackrel{\&RightArrow;}{\mathrm{DS}}((R_{\mathrm{if}}\&CircleTimes;W_i)*E_i)---\left(16\right)$

$\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{E}^i=1,$ $\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{E}_i=1,$ $\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{r}_i=1---\left(17\right)$

Sup is the similarity between the evidence body, introduces the distance function that Jousselme proposes and obtains, and establishes δ and is one and comprise the complete framework of identification of N proposition in twos, and P (δ) is the set that all subclass of δ generate, II _{P (δ)}Be the burnt first vector space of evidence, its base is the element { A among the P (δ) ₁, A ₂..., A _r. if V ∈ II _{P (δ)}, be expressed as

V=[α ₁,α ₂,…,α _r] （18）

In the formula, α _i∈ R, i=1,2 ..., r. is if a basic trust is distributed in II _{P (δ)}In can represent a vector M, M is with M (A _i) be coordinate system,

M=[M(A ₁),M(A ₂),…,M(A _r)] （19）

In the formula, M (A _i)>=0, i=1,2 ..., r,

If M _iAnd M _jBe 2 BPA on the framework of identification δ, can get M _iAnd M _jBetween distance

$d_{\mathrm{ij}}=\sqrt{\frac{1}{2}(M_i-M_j)D(M_i-M_j)}---\left(20\right)$

In the formula, D is one 2 ⁿ* 2 ⁿMatrix,

$D(A_i,A_j)=\frac{|A_i\&cap;A_j|}{|A_i\&cup;A_j|},i,j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,r$

| A _i∩ A _j| be the number .d of element in the set _IjConcrete computing method are

$d_{\mathrm{ij}}=\sqrt{\frac{1}{2}({\left|\right|M_i\left|\right|}^2+{\left|\right|M_j\left|\right|}^2-2\&lang;M_i-M_j\&rang;)}---\left(22\right)$

Wherein, ‖ M ‖ ²=<m, M>,<m _i, M _j>Be the inner product of two vectors,

$\&lang;M_i,M_j\&rang;=\underset{l=1}{\overset{2^n}{\mathrm{\&Sigma;}}}\underset{p=1}{\overset{2^n}{\mathrm{\&Sigma;}}}{M}_i\left(A_l\right)M_j\left(A_p\right)\frac{|A_l\&cap;A_p|}{|A_l\&cap;A_p|}---\left(23\right)$

The evidence number that the system of setting up departments is collected is q, utilizes formula (23) to obtain evidence body M _iAnd M _jBetween the distance of evidence in twos, it is expressed as a distance matrix:

$D_m=\left[\begin{array}{cccccc}0& d_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& d_{1j}& \&CenterDot;\&CenterDot;\&CenterDot;& d_{1q}\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ d_{i1}& {\mathrm{<figure-callout\; id="2"\; label="d\; i"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}}_i& \&CenterDot;\&CenterDot;\&CenterDot;& d_{\mathrm{ij}}& \&CenterDot;\&CenterDot;\&CenterDot;& d_{\mathrm{iq}}\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ d_{q1}& {\mathrm{<figure-callout\; id="2"\; label="d\; q"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">d</figure-callout>}}_q& \&CenterDot;\&CenterDot;\&CenterDot;& d_{\mathrm{qj}}& \&CenterDot;\&CenterDot;\&CenterDot;& 0\end{array}\right]---\left(24\right)$

Definition evidence body M _iAnd M _jBetween similarity measure

S _ij=1-d _ij i,j=1,2,…,q （25）

Its result representes with a similarity matrix:

$S_m=\left(\begin{array}{cccccc}1& S_{12}& \&CenterDot;\&CenterDot;\&CenterDot;& S_{\mathrm{ij}}& \&CenterDot;\&CenterDot;\&CenterDot;& S_{1q}\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ S_{i1}& {\mathrm{<figure-callout\; id="2"\; label="S\; i"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">S</figure-callout>}}_i& \&CenterDot;\&CenterDot;\&CenterDot;& S_{\mathrm{ij}}& \&CenterDot;\&CenterDot;\&CenterDot;& S_{\mathrm{iq}}\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ \&CenterDot;& \&CenterDot;& & \&CenterDot;& & \&CenterDot;\\ S_{q1}& S_{\mathrm{qw}}& \&CenterDot;\&CenterDot;\&CenterDot;& S_{\mathrm{qi}}& \&CenterDot;\&CenterDot;\&CenterDot;& 1\end{array}\right)---\left(26\right)$

Distance between two evidence bodies is more little, and their similarity degree is also just big more. and E is E by the evidence collection _i=(E _I1, E _I2..., E _Ix), E wherein _I1, E _I2..., E _IxFor for i class fault from 1 to x evidence that various information source obtains, define the likelihood similarity so and do

$\left\{\begin{array}{c}{S}_{E^i{E}_i}=1-d_{E^i{E}_i}\\ F=\mathrm{Sup}(E^i,E_i)=S_{E^i{E}_i}\end{array}\right.i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,x---\left(27\right)$

Formula (18)-(27) are explains the specific algorithm of formula (15).

Obviously; Necessarily exist such priori weight factor to make to exist and satisfy formula (15) and obtain maximal value; That is to say that there are very big similarity in the evidence body and the desired result evidence body that make final evidence combination obtain, also have only through trying to achieve likelihood similarity maximal value and can judge that each information source is for " sensitivity " degree that reflects on the different faults.

Through above-mentioned (15)-(27) formula and utilize genetic algorithm to ask for satisfied (15) for peaked coefficient; But this coefficient is exactly the priori weight coefficient about temperature evidence value and measuring point information of voltage fault evidence value, and it is following specifically to utilize genetic algorithm to ask for step:

Utilize genetic algorithm to obtain the priori weight coefficient:, can find global optimization to separate and avoid occurring local convergence, so carry out the design of priori weight factor as fitness function with following formula (15) because genetic algorithm is adapted to finding the solution of all kinds of complicated optimum problem.Its concrete process flow diagram is seen accompanying drawing 4.

The form that the chromosome representative is separated comprises all hereditary information.Adopt real coding among this paper, chromosome can be expressed as the gene strand of two dimension,

$X={[L_0^{\&prime;},C_0]}^T---\left(28\right)$

Based on optimization aim, the structure fitness function is following:

If f _1iBe the functional value of F (X) about chromosome i.In searching process, should make population to f _1iBigger direction is evolved and should be avoided being absorbed in the locally optimal solution of F (X).In view of the above, following to t for the chromosome selection operation:

(1) keeps breeding, in the hands-on, the top n individuality is directly entailed the next generation by the fitness ordering;

(2) microhabitat rule of elimination because niche technique is avoided local convergence and precocity in genetic algorithm, is kept the multifarious a kind of effective method of population.With the Z that obtains in the genetic algorithm ₁It is Z that the N individuals merging that individuals and elite keep obtains the new population number ₁+ N, can obtain 2 individuals X in new population according to formula _iAnd X _jBetween the hamming distance, promptly

$\left|\right|X_i-X_j\left|\right|=\sqrt{\underset{k=1}{\overset{Z}{\mathrm{\&Sigma;}}}{(x_{\mathrm{ik}}-x_{\mathrm{jk}})}^2}---\left(30\right)$

(i=1,2,…,Z ₁+N-1;j=i+1,…,Z ₁+N)

As ‖ X _i-X _j‖<l _hWhen (in circuit test, being taken as 0.025), more individual X _iWith individual X _jFitness size, and the individuality that wherein fitness is lower sentenced penalty, in next round is evolved, eliminate.

(3) following column selection probability is evolved:

$P_S=\frac{f_{1i}}{\underset{k=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="Z"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">Z</figure-callout>}}_1+N}{\mathrm{\&Sigma;}}}{f}_{1k}}(1-\frac{\min \left\{\right|f_{1i}|,|f_{1i}-L\left|\right\}}{\underset{k=1}{\overset{N}{\mathrm{\&Sigma;}}}\min \left\{\right|f_{1k}|,|f_{1k}-L\left|\right\}})---\left(31\right)$

(4) for keeping population scale constant, after above-mentioned selection operation is accomplished, according to selecting chromosome quantity, duplicate the breeding chromosome of some, to supply population.

Be absorbed in local solution for fear of algorithm, adopt a kind of method of self-correcting parameter adjustment to overcome the precocity convergence of algorithm, promptly using genetic algorithm to carry out in the process of parameter optimization, P _CAnd P _MCarry out self-adjusting, let them increase with the chromosome fitness value and diminish, along with the chromosome adaptive value reduces and becomes big, computing formula is following:

Crossover probability:

$P_C=\left\{\begin{array}{c}{K}_1\frac{f_{\max }-f^{\&prime;}}{f_{\max }-f_{\mathrm{avg}}},f^{\&prime;}\&GreaterEqual;{\mathrm{<figure-callout\; id="2"\; label="f\; avg\; K"\; filenames="H2009100424084E00032.png"\; state="\{\{state\}\}">f</figure-callout>}}_{\mathrm{avg}}& \\ K_{}{}_2,f^{\&prime;}\&le;f_{\mathrm{avg}}\end{array}\right.---\left(32\right)$

The variation probability:

$P_M=\left\{\begin{array}{c}{K}_3\frac{f_{\max }-f}{f_{\max }-f_{\mathrm{avg}}},f\&GreaterEqual;f_{\mathrm{avg}}\\ K_4,f\&le;f_{\mathrm{avg}}\end{array}\right.---\left(33\right)$

F in the formula _MaxMaximal value for chromosome fitness in the current population; f _AvgBe the average fitness value of chromosome in the current population; F ' is bigger one of two chromosomal fitness of participating in interlace operation; F is for participating in the chromosome fitness value of variation; K ₁, K ₂, K ₃, K ₄Be to be not more than 1 positive constant, can be by the corresponding adjustment of particular problem.

By above analysis, the algorithm steps that is optimized is following:

(1) according to requirement of system design initialization constant, puts t=0;

(2) the generation scale is the initial population of N in domain;

(3) according to above-mentioned choice mechanism chromosome is selected;

(4) with probability P _cChromosome is carried out allele to intersect;

(5) with probability P _MChromosome is made a variation;

(6) chromosome fitness in the population is assessed, if all chromosomes all satisfy two constraint conditions in the population, and the maximum f of continuous three generations population ₁Variable quantity all less than a very little positive number, then algorithm is restrained; If algorithm is not convergence as yet, then t=t+1 forwards (3) to.

3 synthesis steps

Can know by last two joints, but all obtain that then this synthetic method is following about the priori weight factor and the associated weights coefficient of temperature evidence value and measuring point information of voltage fault evidence value:

But step 1 is framework of identification with the temperature evidence value that obtains with measuring point information of voltage fault evidence value, but obtains the associated weights coefficient w of evidence source temperature evidence value and measuring point information of voltage fault evidence value _f=(w ₁, w ₂..., w _x) and priori weight coefficient R _If=(r _I1, r _I2..., r _Ix) (i=1,2 ..., x).

Step 2 is asked for about the unified weight allocation coefficient of these two types of evidence values following:

${\mathrm{\&xi;}}_{\mathrm{\&kappa;j}}=\frac{w_{\mathrm{\&kappa;}}+r_{\mathrm{\&kappa;j}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_{\mathrm{ij}}+\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i},j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,x---\left(34\right)$

$\mathrm{st}.\underset{\mathrm{\&kappa;}=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&xi;}}_{\mathrm{\&kappa;j}}=1---\left(35\right)$

The number of priori weight coefficient, associated weights coefficient and unified weight coefficient is x, ξ in the formula _{κ j}Be illustrated in the j class fault that adds up in the m class and be the unified weight allocation coefficient under the κ class at sensor. for this method, k gets 1.2 two types of evidence values.

Step 3 is utilized the elementary probability apportioning cost of following formula as all propositions in each framework of identification of adjustment, and then adjustment back elementary probability apportioning cost is:

$m_j^{*}\left(A_i\right)={\mathrm{\&xi;}}_{\mathrm{\&kappa;j}}m\left(A_i\right)---\left(36\right)$

In the formula: system exists k evidence source (k=1 or 2) and m class fault, j=1 then, and 2 ..., x, j represent j class fault.

Though step 4 can obtain the unified weight allocation coefficient of m group by (34), can occur a problem when being applied to (36) to this, the m (A that is promptly obtained by sensor _i) can not judge that in advance which belongs to organizes fault, therefore will obtain adjusted elementary probability apportioning cost must at first confirm how to choose ξ _{κ j}In the k class value.But must $\&Exists;{\mathrm{\&xi;}}_{\mathrm{\&kappa;\; j}},{\mathrm{\&xi;}}_{\mathrm{\&kappa;\; j}}\&Element;\mathrm{\&gamma;},\mathrm{\&gamma;}=\left\{{\mathrm{\&xi;}}_{\mathrm{\&kappa;\; j}}\right|{\mathrm{\&xi;}}_{11},{\mathrm{\&xi;}}_{12},\&CenterDot;\&CenterDot;\&CenterDot;,{\mathrm{\&xi;}}_{1n},{\mathrm{\&xi;}}_{21},\&CenterDot;\&CenterDot;\&CenterDot;{\mathrm{\&xi;}}_{\mathrm{Mn}}\},$ Make in (40) $m_j^{*}\left(A_i\right)(j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,n)$ The likelihood similarity

Maximum, that is:

(1) calculates m group elementary probability apportioning cost

(2) calculate this m group elementary probability apportioning cost

Under m group likelihood similarity sup (m ^* _j(A _i), E ⁱ)

(3) find out the ξ that makes that the likelihood similarity is maximum _{κ j}

After step 5 obtains adjusted elementary probability apportioning cost ; It is not 1, replenish following formula for this reason

$m_j^{*}\left(\mathrm{\&theta;}\right)=1-\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{m}_j^{*}\left(A_i\right)---\left(37\right)$

Step 6 is brought all adjusted basic allocation probabilitys of assigning a topic into the D-S formula at last.

The D-S evidence theory provides a very useful composite formula, can synthesize the evidence in a plurality of evidences source.Formula definition is following

In the formula: k _oSize reaction the total conflict spectrum between on evidence.1-k _oBe called normalized factor, its effect is for fear of when synthetic, the probability of non-zero being composed to empty set.m _i(B _i) be the evidence value that obtains after but the above-mentioned temperature evidence value that obtains is adjusted through unified weight coefficient with measuring point information of voltage fault evidence value, m (B) is the evidence value after merging through D-S, makes a strategic decision through D-S evidence formula, obtains localization of fault.

The present invention has at first set up the extracting rule of temperature fault information; The mathematical model of membership function, judgment condition and the recognition methods of temperature fault information has been proposed; After on the basis of D-S evidence theory, having analyzed the problem that occurs when utilizing the D-S evidence theory to merge conflicting information; Proposed to have improved the Circuits System accuracy of fault diagnosis based on the new fusion method of associated weights coefficient with the weight coefficient that conflicts.

Embodiment

We choose low-pass filter circuit shown in Figure 5 and diagnose.The nominal value and the tolerance of each element of wave filter are seen table 1, and are chosen under 21 excitation frequencies and take a sample, and in order to extract fault sample, we are from definition [0.1Xn at interval; (1-t) Xn] and [(1+t) Xn; 10Xn] between same distribution in obtain the fault element value, t represent range of tolerable variance here, Xn represents the nominal value of circuit component, white noise and its fault that we have added 30db to the output of circuit are chosen and are component-level.

When the R1 soft fault, its resistance is following table 2 less than nominal value but can get measuring point voltage; Table 3 has provided in the i.e. second type of evidence value of the temperature fault degree of membership value of extracting through actual test back, and (the evidence value is represented respectively for R1, R2 through temperature evidence value that neural network N1 exports; C1, the probability of malfunction value of C2 i.e. [R1, R2; C1, C2]; Table 4 has provided based on associated weights coefficient of the method and priori weight coefficient; Table 5 has provided unified weight coefficient and fusion results.Can find out R1 fault evidence value up to 0.9148 from the result, localization of fault is accurate and higher than single evidence value.

Table 1

Element	Nominal value	Tolerance
			C 1	50nF	±5%
C 2	50nF	±5%
			R 1	100Ω	±10%
R 2	100Ω	±10%
			R A	5KΩ	±1%
R B	2KΩ	±1%

Table 2

Table 3

Table 4

The evidence value	Collision vector	The normalization collision vector	The associated weights coefficient	The priori weight coefficient
					[0.90,0.10,0,0]	[0.3668,0.1980]	[0.6494,0.3506]	0.3397	0.4223
[0.69,0.13,0.07,0.11]	[0.3668,0.3238]	[0.5311,0.4689]	0.3192	0.1847
					[0.89,0.01,0.09,0.01]	[0.1980,0.3238]	[0.3795,0.6205]	0.3411	0.3930

Table 5

Claims (6)

1. analog-circuit fault diagnosis method that merges based on heterogeneous information may further comprise the steps:

1) utilize temperature sensor measurement to treat the temperature change value of diagnostic circuit element;

2) ask for the optimum gate threshold values of each element temperature change value, its step is following:

A1) the component temperature changing value information of obtaining according to sensor utilizes iterative formula to calculate family of power and influence's value;

A2) family of power and influence's value that utilize to obtain is cut apart and is handled each element temperature change value; Through the temperature change value information calculations optimized discrete degree that obtains after the dividing processing and calculate element number,, the element number principle that satisfies defined and optimized discrete stop calculating when spending;

A3) then do not carry out next step iterative computation if do not meet the demands, repeat above-mentioned steps a2) and a3), till obtaining the best family of power and influence's value that satisfies condition, thereby obtain the optimum gate threshold values;

3) the optimum gate threshold values filtering element temperature change value that utilize to obtain, when certain element temperature change value then is included into it among failure domain Φ during greater than the optimum gate threshold values, otherwise this temperature variation value information of filtering;

4) temperature change value with failure domain Φ is input in the temperature degree of membership model, utilizes membership function calculate to obtain the fault degree of membership value of each element temperature change value, with it as temperature evidence value;

5) treat diagnostic circuit and apply pumping signal, but measure measuring point voltage;

6) but measuring point voltage input BP neural network is carried out elementary diagnosis, but obtain the evidence value of measuring point voltage, and with it as voltage evidence value;

7) utilize the heterogeneous information emerging system that said temperature evidence value and voltage evidence value are merged, confirm fault element, the steps include:

The evidence collision vector of a, accounting temperature evidence value and voltage evidence value with the normalization of evidence collision vector and ask entropy, is asked the associated weights coefficient that obtains reciprocal again;

B, temperature evidence value, voltage evidence value are compared with desirable fault output vector respectively, calculate both likelihood similarities;

C, according to each group likelihood similarity utilize genetic algorithm to seek to satisfy similarity maximum respectively organize the priori weight coefficient;

D, go out unified weight allocation coefficient according to priori weight coefficient and associated weights coefficient calculations, unified weight coefficient computing formula does

${\mathrm{\&xi;}}_{\mathrm{\&kappa;j}}=\frac{w_{\mathrm{\&kappa;}}+r_{\mathrm{\&kappa;j}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_{\mathrm{ij}}+\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_i},j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,x$

$\mathrm{st}.\underset{\mathrm{\&kappa;}=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&xi;}}_{\mathrm{\&kappa;j}}=1$

The number of priori weight coefficient, associated weights coefficient and unified weight coefficient is x, ξ in the formula _{κ j}Be illustrated in the j class fault that adds up in the m class and be the unified weight allocation coefficient under the κ class at sensor, but accounting temperature evidence primary system one weight coefficient during with measuring point voltage evidence primary system one weight coefficient k get 1 and 2 respectively, w _κFor being the associated weights coefficient under the κ class at sensor, r _KjFor being that the κ class is the priori weight coefficient in j class fault and at sensor;

E, the unified weight allocation coefficient of utilization are readjusted each evidence value, carry out D-S again and merge the fault location element.

2. the analog-circuit fault diagnosis method that merges based on heterogeneous information according to claim 1, the membership function in the said step 4) is:

<math> <mrow> <msub> <mi>F</mi> <mi>ij</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open='{' close=''> <mtable> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mfrac> <mrow> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mrow> </mfrac> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <msup> <mrow> <mo>(</mo> <mfrac> <msub> <mi>x</mi> <mi>j</mi> </msub> <mi>&amp;lambda;&amp;gamma;</mi> </mfrac> <mo>)</mo> </mrow> <mi>&amp;gamma;</mi> </msup> <mi>exp</mi> <mrow> <mo>(</mo> <mi>&amp;gamma;</mi> <mo>-</mo> <mfrac> <msub> <mi>x</mi> <mi>j</mi> </msub> <mi>&amp;lambda;</mi> </mfrac> <mo>)</mo> </mrow> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mo>-</mo> <mfrac> <mrow> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mi>ij</mi> </msub> <mo>-</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> </mrow> </mfrac> </mtd> <mtd> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>e</mi> <mi>ij</mi> </msub> <mo>&lt;;</mo> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>&amp;le;</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> <mtr> <mtd> <mn>0</mn> </mtd> <mtd> <msub> <mi>x</mi> <mi>j</mi> </msub> <mo>></mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>+</mo> <msub> <mi>t</mi> <mi>ij</mi> </msub> </mtd> </mtr> </mtable> </mfenced> </mrow></math>

Wherein, x ₀Be the circuit working normal parameter value of detected element just often; e _IjFor waiting to diagnose the normal variation scope of component parameters; t _IjFor waiting to diagnose the limit deviation of component parameters; F _IjFor sensor j measures the fault degree of membership that measurand belongs to the i pattern; x _jCircuit component temperature change value and x for j class sensor test _j∈ Φ; And λ γ=max (Δ t ₁, Δ t ₂..., Δ t _m), Δ t ₁, Δ t ₂..., Δ t _mThe temperature change value of expression fault element.

3. the family of power and influence the analog-circuit fault diagnosis method that merges based on heterogeneous information according to claim 1, said step a1) is worth iterative formula and is:

ε ₁=|T/card(Θ)-max{Δt|Δt ₁,Δt ₂，…,Δt _m}|+T/card(Θ)

${\mathrm{\&epsiv;}}_2=\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{t}_1,\mathrm{\&Delta;}{t}_2,\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{\&Delta;}{t}_m\}-{\mathrm{\&epsiv;}}_1|}{2}+{\mathrm{\&epsiv;}}_1$

.

.

.

.

.

.

${\mathrm{\&epsiv;}}_h=\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{t}_1,\mathrm{\&Delta;}{t}_2,\&CenterDot;\&CenterDot;\&CenterDot;\mathrm{\&Delta;}{t}_m\}-{\mathrm{\&epsiv;}}_{h-1}|}{2}+{\mathrm{\&epsiv;}}_{h-1}$

ε in the above-mentioned formula _hRepresent the family of power and influence's numerical value after the iteration the h time, Θ is the ensemble of communication of fault element temperature change value, the number of element among card (Θ) the expression set Θ; N is original temperature variation value information number; M is for being worth the temperature change value number after iteration is screened, card (Θ)=n through the family of power and influence

Δ t _iTemperature change value for fault element.

4. the analog-circuit fault diagnosis method that merges based on heterogeneous information according to claim 3, said iterative formula judgment condition is:

1.. when the ε that obtains can make that element number is not more than 3 in failure domain Φ, and when the arbitrary temp changing value all is not less than the family of power and influence and is worth among the failure domain Φ, termination of iterations;

2.. cut apart the failure domain Φ that obtains by ε and should satisfy following formula: promptly the ratio of element number is less than L among the difference of maximum temperature changing value and minimum temperature changing value and the failure domain Φ, and wherein L gets 30-50,

$\frac{\left|\max \right\{\mathrm{\&Delta;t}|\mathrm{\&Delta;}{t}_1,\mathrm{\&Delta;}{t}_2,\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{\&Delta;}{t}_m\}-\min \left\{\mathrm{\&Delta;t}\right|\mathrm{\&Delta;}{t}_1,\mathrm{\&Delta;}{t}_2,\&CenterDot;\&CenterDot;\&CenterDot;,\mathrm{\&Delta;}{t}_m\left\}\right|}{\mathrm{card}\left(\mathrm{\&Phi;}\right)}\&le;L$

Δ t ₁, Δ t ₂..., Δ t _mThe temperature change value of expression fault element, the number of element among card (Φ) the expression failure domain Φ, Φ is the set that the fault element temperature change value obtains after family of power and influence's value is cut apart filtration.

5. the analog-circuit fault diagnosis method that merges based on heterogeneous information according to claim 1, the likelihood similarity formula among the said step b is:

$\left\{\begin{array}{c}{S}_{E^i{E}_i}=1-d_{E^i{E}_i}\\ F=\mathrm{Sup}(E^i,E_i)=S_{E^i{E}_i}\end{array}\right.i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;,x$

Wherein, S _IjBe temperature evidence value M _iBut with measuring point voltage evidence value M _jBetween similarity measure, Sup is the similarity between the evidence body, E _iTemperature evidence value and voltage evidence value are for being the evidential probability on the i class fault, E ⁱBe the desirable evidential probability about i class fault, d _IjBe temperature evidence value M _iBut with measuring point voltage evidence value M _jBetween distance,

Expression evidence body E _iAnd E ⁱBetween similarity measure; Expression evidence body E _iAnd E ⁱBetween distance.

6. the analog-circuit fault diagnosis method that merges based on heterogeneous information according to claim 1, the maximum priori weight coefficient of respectively organizing of similarity that satisfies among the said step c is asked for formula and is:

Max：F=Sup(E ⁱ,E ^* _i)

$s.t{E^{*}}_i=\stackrel{\&RightArrow;}{\mathrm{DS}}((R_{\mathrm{if}}\&CircleTimes;W_i)*E_i)$

$\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{E}^i=1,$ $\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{E}_i=1,$ $\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{r}_i=1$

E wherein _iTemperature evidence value and voltage evidence value are for being the evidential probability on the i class fault, E ⁱBe desirable evidential probability about i class fault,

Be D-S blending algorithm operator,

For merging with the D-S amalgamation mode, Sup is the similarity between the evidence body, r _iBe the list value of relative priori weight, R _IfBe the priori weight coefficient.

Images