CN104102215A - Transformer and transformer peripheral circuit fault control system - Google Patents

Abstract

The invention discloses a transformer and transformer peripheral circuit fault control system. A neural network system suitable for latent fault control of the transformer is invented according to the characteristics of transformer fault control and by analyzing several common fault control standards and methods; and fuzzy theory knowledge is introduced to deduce the fuzzy relation between the fault and omens, and primary diagnosis on the fault positions of the transformer is carried out.

Description

Transformer and peripheral circuit fault control system thereof

Technical field

The present invention relates to a kind of mechanical fault control system, particularly a kind of transformer and peripheral circuit fault control system thereof.

Background technology

Power transformer is converting equipment important in electric system, and its running status directly affects security, the stability of system.Finding early the latency fault of transformer, guarantee the safe operation of transformer, is the major issue that power department is paid close attention to thereby improve the reliability of powering.Facts have proved that the electric method of grinding of assisting of three-ratio method and improvement is very effective for latency fault and the development trend thereof of finding transformer inside.Method is ground by three-ratio method and improvement electricity association, because of simple, convenient, in various countries, is widely used.But, due to the difference of transformer capacity, electric pressure, insulation system, history run, manufacturer and system of living in, cause its accuracy generally can only reach 80% left and right.

Transformer fault is the result of transformer itself and applied environment combined action and long-term accumulation, thereby the sign of transformer fault is varied, contacting between one failure symptom and failure mechanism is also intricate, and this has just caused very large difficulty to setting up general transformer fault control method.Up to now, the research of transformer fault control technology is perfect not enough, particularly aspect being effectively combined of used analytical approach and concrete diagnostic field knowledge, also has problems.

Artificial neural network is a kind of intelligentized data processing method, and its processing has the ability of nonlinear relationship data, is that other method is incomparable at present.The distributed store of its input and output Nonlinear Mapping characteristic, information, parallel processing and the effect of overall collective, particularly self-organization and self-learning capability of its height, become a kind of effective ways and the means of fail-safe control.It provides a kind of brand-new knowledge acquisition, knowledge representation and knowledge reasoning mode, for fail-safe control problem provides a kind of new solution route.Use it in the fail-safe control of complication system, can solve preferably " bottleneck " problem of conventional fault expert system for control knowledge acquisition.

Study and show both at home and abroad, in fail-safe control field, the accuracy rate of Neural Network Diagnosis Method can reach more than 80%, has potential practical value.

For this reason, the feature that the present invention controls according to Power Transformer Faults, has worked out the neural network comprehensive diagnos system that is suitable for latent transformer fail-safe control.

Summary of the invention

Technical matters to be solved by this invention is to be: use IEC three-ratio method to carry out fail-safe control analysis to the running status of transformer, because process is complicated, calculate loaded down with trivial detailsly, the reasons such as ratio interval of ratio coding obscurity boundary usually cause erroneous judgement.Due to the deficiency of IEC three-ratio method itself, even sometimes also there will be the situation that can not provide diagnosis.

For IEC technology has been given play to larger effect in the fail-safe control of transformer; improve the accuracy of IEC three-ratio method; the present invention utilizes visualization technique to carry out graphic to three ratios, from the angle of multidimensional, gives data space attribute, and sets up the visual fail-safe control model of transformer; Utilize Vague Diagnosis Technique that failure symptom signal and fuzzy mathematics method are combined, the blooming subset of gas composition ratio being regarded as to fault category, according to the mutual ratio of the gas with various component providing in three-ratio method and the quantitative relationship between transformer fault classification.

Set up blooming subset corresponding to the subordinate function of different faults classification, thereby carry out the fuzzy diagnosis of fault category; Adopt integrated method, visualization technique and Vague Diagnosis Technique are combined and effectively reduce that the inevitable diagnosis bringing while adopting single transformer fault control method to diagnose makes mistakes or the diagnostic result erroneous judgement problem such as comprehensively not.

In order to solve the problems of the technologies described above, the invention provides a kind of transformer and peripheral circuit fault control method thereof, comprising: signal detection step, feature extraction step, state recognition step and forecast steps in decision-making;

Described signal detection step, for measuring the status signal relevant with fault; Its signal detection process comprises following aspect: 1) signal measures process: the sniffer consisting of electric weight or non-electrical sensor obtains the transformation of variable of measuring object ginseng; 2) signal condition process: the signal that sniffer is obtained converts and transmits; 3) data acquisition: signal from becoming continuously discrete process; According to different diagnostic purposes and object, select can reflect the signal of measurand state, make sensing device, signal regulating device and the data collector of power consumption or non electrical quantity, image data is arranged, form the status information storehouse of measurand;

Described feature extraction step, for the signal conversion detecting is carried out to feature extraction, by signal analysis and processing, extracts the feature of energy faults state by the initial state signal being obtained by input, forms pattern to be checked; Aforementioned signal analysis and processing comprises: remove noise processed, amplification or compression processing, signal conversion.。

Described state recognition step, for analysis of failure case, utilize database technology to set up fault file storehouse, the state model in fault file storehouse is decided to be to reference mode, calling pattern to be checked by the state model obtaining after input, feature extraction after kainogenesis fault, and compare judgement with reference mode, normally whether discriminating device operation.

Described forecast steps in decision-making, for through differentiating, detects belonging to the continuation of normal condition, repeats above operation; To belonging to abnormality, find out failure condition, make trend analysis, estimate Future Development and the time of continuing operation, and propose control measure and maintenance decision according to failure cause;

Described signal detection step further comprises that transient state equivalent circuit detects; Described transient state equivalent circuit detects, and is further used for analyzing transient state capacitance current; Described transient state equivalent circuit is when analyzing transient state capacitance current, and rL and L do not consider, and the series loop and the zero sequence sinusoidal voltage u0 that utilize L0, C, R0 to form calculate transient state capacitance current according to following differential equation:

In formula: the amplitude that Um is residual voltage;

C represents the zero sequence equivalent capacity of electrical network;

L ₀for the zero sequence equivalent inductance containing transformer and line inductance;

R ₀represent the substitutional resistance in zero sequence loop;

R _l, L represents respectively active loss resistance and the inductance of arc suppression coil; u ₀for residual voltage.

In the inner detection of described power transformer, utilize 5 kinds of characteristic gas C ₂h ₄, C ₂h ₆, C ₂h ₂, CH ₄, H ₂form three correlative value C ₂h ₂/ C ₂h ₄, CH ₄/ H ₂and C ₂h ₄/ C ₂h ₆, identical in the situation that these ratios with different coded representations, as shown in table 1 below, according to test result calculations, draw coding, and three correlative values be converted into corresponding code set, the correspondence of then tabling look-up draws fault type and trouble location, as shown in table 2 below:

Table 1 coding rule

Table 2 fault type determination methods

For solving the problems of the technologies described above, the present invention also provides a kind of transformer fault control method, comprising: power transformer peripheral circuit detects submethod, and, the inner submethod that detects of power transformer;

Described power transformer is inner to be detected in submethod, CH ₄/ H ₂the make of ratio coding subordinate function is: CH ₄/ H ₂the Fuzzy Distribution function that is under the jurisdiction of respectively 0,1,2 codings:

$v_0\left(x\right)=\left\{\begin{array}{cc}0,& x<0.08\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 1,& 0.12<x\&le;0.9\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-1.0),& 0.9<x\&le;1.1\\ 0,& 1.1<x\end{array}\right.$

$v_1\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.08\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 0,& 0.12<x\end{array}\right.$

$v_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;0.9\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-1)& 0.9<x\&le;1.1\\ 1,& 1.1<x\end{array}\right.$

The inner submethod that detects of described power transformer, further comprises neural network failure control step; The inner detection subsystem of described power transformer is input to the sample set of output by described neural network learning, realization is input to the mapping relations of output, the Nonlinear Mapping relation of phenomenon of the failure and reason can be when thering is minimum computing, can set up, the pattern of arbitrary continuation can be stored to (X _k, T _k; K=1,2 ..., m), complete input { X _kto desired output (Z _kmapping.

For solving the problems of the technologies described above, the present invention provides again a kind of transformer and peripheral circuit fault control system thereof, comprising: signal detecting part, feature extraction parts, state recognition parts and forecast decision component;

Described signal detecting part, for measuring the status signal relevant with fault; Its signal detection process comprises following aspect: 1) signal measures process: the sniffer consisting of electric weight or non-electrical sensor obtains the transformation of variable of measuring object ginseng; 2) signal condition process: the signal that sniffer is obtained converts and transmits; 3) data acquisition: signal from becoming continuously discrete process; According to different diagnostic purposes and object, select can reflect the signal of measurand state, make sensing device, signal regulating device and the data collector of power consumption or non electrical quantity, image data is arranged, form the status information storehouse of measurand;

Described feature extraction parts, for the signal conversion detecting is carried out to feature extraction, by signal analysis and processing, extract the feature of energy faults state by the initial state signal being obtained by input, form pattern to be checked; Aforementioned signal analysis and processing comprises: remove noise processed, amplification or compression processing, signal conversion.

Described state recognition parts, for analysis of failure case, utilize database technology to set up fault file storehouse, the state model in fault file storehouse is decided to be to reference mode, calling pattern to be checked by the state model obtaining after input, feature extraction after kainogenesis fault, and compare judgement with reference mode, normally whether discriminating device operation.

Described forecast decision component, for through differentiating, detects belonging to the continuation of normal condition, repeats above operation; To belonging to abnormality, find out failure condition, make trend analysis, estimate Future Development and the time of continuing operation, and propose control measure and maintenance decision according to failure cause;

Described signal detecting part further comprises transient state equivalent circuit; Described transient state equivalent circuit, is further used for analyzing transient state capacitance current; Described transient state equivalent circuit when analyzing transient state capacitance current, r _ldo not consider with L, utilize L ₀, C, R ₀the series loop and the zero sequence sinusoidal voltage u that form ₀, according to following differential equation, calculate transient state capacitance current:

In formula: U _mamplitude for residual voltage;

C represents the zero sequence equivalent capacity of electrical network;

L ₀for the zero sequence equivalent inductance containing transformer and line inductance;

R ₀represent the substitutional resistance in zero sequence loop;

R _l, L represents respectively active loss resistance and the inductance of arc suppression coil; u ₀for residual voltage.

Described transformer and peripheral circuit fault control system thereof, utilize 5 kinds of characteristic gas C ₂h ₄, C ₂h ₆, C ₂h ₂, CH ₄, H ₂form three correlative value C ₂h ₂/ C ₂h ₄, CH ₄/ H ₂and C ₂h ₄/ C ₂h ₆, identical in the situation that these ratios with different coded representations, as shown in table 1 below, according to test result calculations, draw coding, and three correlative values be converted into corresponding code set, the correspondence of then tabling look-up draws fault type and trouble location, as shown in table 2 below:

Table 1 coding rule

Table 2 fault type determination methods

For solving the problems of the technologies described above, the present invention provides a kind of transformer fault control system again, comprising: power transformer peripheral circuit detection subsystem, and, the inner detection subsystem of power transformer;

In the inner detection subsystem of described power transformer, CH ₄/ H ₂the make of ratio coding subordinate function is: CH ₄/ H ₂the Fuzzy Distribution function that is under the jurisdiction of respectively 0,1,2 codings:

$v_0\left(x\right)=\left\{\begin{array}{cc}0,& x<0.08\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 1,& 0.12<x\&le;0.9\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-1.0),& 0.9<x\&le;1.1\\ 0,& 1.1<x\end{array}\right.$

$v_1\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.08\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 0,& 0.12<x\end{array}\right.$

$v_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;0.9\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-1)& 0.9<x\&le;1.1\\ 1,& 1.1<x\end{array}\right.$

The inner detection subsystem of described power transformer, further comprises neural network failure control assembly; The inner detection subsystem of described power transformer is input to the sample set of output by described neural network learning, realization is input to the mapping relations of output, the Nonlinear Mapping relation of phenomenon of the failure and reason can be when thering is minimum computing, can set up, the pattern of arbitrary continuation can be stored to (X _k, T _k; K=l, 2 ..., m), complete input { X _kto desired output { Z _kmapping.

The technique effect that the present invention is useful is: for describing the various characteristics of transformer fault and effectively analyzing, learn from other's strong points to offset one's weaknesses the artificial intelligence technologys such as neural network, fuzzy theory, pattern-recognition are fused into one, form the systems approach of transformer fault control analysis.For the deficiency of IEC three-ratio method in prior art, adopt visualization technique, the visual fail-safe control model of research transformer, and utilize Vague Diagnosis Technique to carry out the Fuzzy processing of ratio coding.

Adopt integrated method, visualization technique and Vague Diagnosis Technique are combined, set up the transformer insulated fault visual Fuzzy Diagnostic System based on IEC tri-ratios.The diagnostic method of the latent transformer fault based on neural network is provided.

The feature of controlling according to oil-filled Power Transformer Faults, analyzes several conventional fail-safe control standard and methods, has invented the nerve network system that is suitable for latent transformer fail-safe control.Introduce fuzzy theory knowledge, the fuzzy relation between derivation fault and sign, carries out the tentative diagnosis at transformer fault position.

Fuzzy diagnosis method and neural network are combined, provide and be suitable for the Fuzzy Neural Network System that transformer fault is controlled, carry out the definite research in transformer fault position.

Accompanying drawing explanation

Fig. 1 is the working plan that power transformer peripheral circuit shown in the embodiment of the present invention detects;

Fig. 2 is the transient state equivalent circuit figure of neutral by arc extinction coil grounding system shown in the embodiment of the present invention;

Fig. 3 is the mapping of neural network shown in embodiment of the present invention mechanism figure;

Fig. 4 is the figure of BP Principles of Network shown in the embodiment of the present invention;

Fig. 5 is the process flow diagram flow chart of BP network learning and training shown in the embodiment of the present invention;

Fig. 6 improves the changing trend diagram of Rogers standard population error shown in the embodiment of the present invention;

Fig. 7 is the neural metwork training result figure of IEC standard shown in the embodiment of the present invention;

Fig. 8 is the Neural Network Diagnosis efficiency diagram of two kinds of diagnostic systems shown in the embodiment of the present invention or method;

Fig. 9 is encoded to 0 characteristic function of a set C described in the embodiment of the present invention ₀(x) figure;

Figure 10 is encoded to 1 characteristic function of a set C described in the embodiment of the present invention ₀(x) figure;

Figure 11 falls ridge shape distribution plan described in the embodiment of the present invention;

Figure 12 is Fuzzy Distribution ridge shape distribution schematic diagram described in the embodiment of the present invention.

Embodiment

Below with reference to drawings and Examples, describe embodiments of the present invention in detail, to the present invention, how application technology means solve technical matters whereby, and the implementation procedure of reaching technique effect can fully understand and implement according to this.

The working routine that power transformer peripheral circuit of the present invention detects.As shown in Figure 1, it comprises the foundation in status information storehouse and fault file storehouse, and four working routines such as input, feature extraction, state recognition and forecast decision-making.

1. input

Mainly to measure the status signal relevant with fault.Status signal is unique carrier of failure message, is also unique foundation of diagnosis.Therefore it is very important in status monitoring, obtaining in time, exactly detected object status information.Signal detection process comprises following aspect: 1) signal measures process: be mainly that the sniffer consisting of electric weight or non-electrical sensor obtains the transformation of variable of measuring object ginseng; 2) signal condition process: Main Function is that the signal that sniffer is obtained converts and transmits; 3) data acquisition: signal from becoming continuously discrete process.Acquisition can accurately reflect that the data of equipment running status are steps of the most critical of fail-safe control, is also that fault is controlled to the key point losing.According to different diagnostic purposes and object, select can reflect the signal of measurand state, the technological means such as sensing device, signal regulating device and data collector that make power consumption or non electrical quantity, arrange image data, form the status information storehouse of measurand.

2. feature extraction

Measurand is often worked in noise and multiple electromagnetic interference environment, so the signal being detected by sniffer also comprises a lot of noise signals and redundant information, these signals can not directly be used for the state characteristic of analytical equipment, must carry out feature extraction through signal conversion.The initial state signal being obtained by input, by signal analysis and processing, as removed the processes such as noise processed, amplification or compression processing, signal conversion, is extracted to the feature of energy faults state, form pattern to be checked.

3. state recognition

Analysis of failure case, utilize database technology to set up fault file storehouse, the state model in fault file storehouse is decided to be to reference mode, calling pattern to be checked by the state model obtaining after input, feature extraction after kainogenesis fault, and compare judgement with reference mode, can discriminating device be normally move or abnormal conditions have occurred.Due to the diversity of fault type, fault vectors pattern is also a lot, and the process of setting up pattern is also to set up fault file.According to the further analysis of failure sign of pattern and state, understand character, the type of fault and produce reason.

4. forecast that decision-making, through differentiating, to belonging to the continued detection of normal condition, repeats above program; To belonging to abnormality, to find out failure condition, make trend analysis, estimate Future Development and the time that can continue operation, and propose control measure and maintenance decision according to failure cause.

For voltage of power device line fault is controlled, current method is mainly to study for single-phase earthing and phase fault, and for high resistive fault and intermittent defect, because after its fault, fault current is little, voltage drop is low, three-phase imbalance feature is not obvious, cause and extract fault signature difficulty relatively.Fault-signal in power transformer system is non-stationary, nonlinear properties often, and for this class signal, time frequency analysis is effective method normally.

Traditional method can not reflect the time variation of non-stationary signal statistic, can not meet the power failure input of non-stationary.Wavelet transformation is owing to having the good reputation of " school microscop " and the characteristic of multiresolution is gathered around and is widely used at aspects such as electric power system fault time detecting, but wavelet analysis method has its shortcoming, one of them is exactly there is no perfect theory in the selection of wavelet basis, just by rule of thumb or attempt, select different wavelet basiss very large to the analyzing influence of result, moreover selected after wavelet basis, whole analytic process all will be used same wavelet basis, cannot change, not there is adaptivity; In addition, the selection of decomposition scale also has certain influence to the result of wavelet analysis, once decomposition scale is selected, decomposition result is only relevant with the analysis frequency of signal, and irrelevant with signal itself, this point, wavelet transformation does not possess adaptive signal decomposition characteristic.

The embodiment of the present invention adopts a kind of new signal processing method, and concrete scheme is as follows:

1. fault detect

After front-end collection system with installation and operation, gather the real time data of power transformer peripheral circuitry, i.e. the three-phase voltage of circuit and three-phase current, by the residual voltage signal of the synthetic three-phase line of three-phase voltage information.In this city strength, the basic frequency that gathers power transformer peripheral circuitry electric network is 50Hz, with fixed sampling frequency, voltage, current signal is sampled simultaneously, obtain voltage sample sequence and current sample sequence, sampling rate is 2000Hz,, within a cycle of operation, samples 40 times.Work as line failure, by processing residual voltage signal, calculate the instantaneous amplitude of the residual voltage single order IMF of three-phase line, then calculate its sampled value sum in a cycle, and compare with predefined threshold value, if be greater than the threshold value of setting, distribution line breaks down, thereby starts failure modes recognizer.

Fault feature vector is implemented dimensionality reduction, extract the major component of fault feature vector, because different faults can produce different principal component patterns, thus the type of definite fault, and failure judgement is singlephase earth fault, two-phase short-circuit fault, line to line fault earth fault or three-phase fault etc.

3. localization of fault

The localization of fault of the embodiment of the present invention refers to circuit branch location, is different from traditional localization of fault, and traditional localization of fault is after line failure, by certain algorithm, locates the position on On-line Fault road; The faulty line judgement of the embodiment of the present invention is also different from traditional failure line selection, traditional failure line selection is that a power transformer peripheral circuit has many outlets, when circuit, there is the faults such as ground connection or short circuit, after the situation of system safety operation of threatening, by certain method or means, determine it is which bar outlet is broken down.Localization of fault in the embodiment of the present invention can only navigate to circuit branch, can not accurately locate fault distance.Specifically refer to according to after being installed on information analysis that the sensing device of power transformer peripheral circuit collects and processing, the fault may or having occurred in circuit is positioned, to the personnel of control center, indicate the general circuit breaking down.The personnel of control center can monitor these circuits by emphasis, according to corresponding fault mode, take certain measure, in case the more generation of catastrophic failure.

In faulty line Position Research, residual voltage and zero sequence current signal, as input signal, calculate zero sequence instantaneous power.In a typical power transformer peripheral circuit zero sequence composition, the characteristic of power direction is:

1) capacitive branch discharges reactive power;

2) perceptual branch absorbing reactive power;

3) resistive branch absorbs active power.

Herein zero sequence instantaneous power theory is located for power transformer peripheral circuit faulty line, utilize each monitoring point place's residual voltage of power transformer peripheral circuit circuit and zero-sequence current to calculate zero sequence instantaneous power, with the monitoring point circuit of the zero sequence instantaneous power opposite direction of regular link be the circuit of failure flow warp, and then according to the actual circuit breaking down of topological structure judgement of distribution system.

After power transformer peripheral circuit breaks down, the numerical value of steady-state component is very little, is easy to be submerged in a large amount of noise signals, has caused the difficulty of extracting useful signal, so often can not correctly judge line fault according to steady-state component.But transient state component amplitude when power transformer peripheral circuit fault occurs is very large, has also comprised abundant failure message.Therefore, the embodiment of the present invention is carried out transient current analysis.

As shown in Figure 2, in the transient state equivalent circuit of neutral by arc extinction coil grounding system, because the analytic process fundamental sum neutral by arc extinction coil grounding system of isolated neutral system is the same, transient state Equivalent Circuit is only equivalent to the r in Fig. 2 _lopen a way with L.

In figure, C represents the zero sequence equivalent capacity of electrical network; L ₀for the zero sequence equivalent inductance containing transformer and line inductance; R ₀represent the substitutional resistance in zero sequence loop; r _l, L represents respectively active loss resistance and the inductance of arc suppression coil; u ₀for residual voltage.

1. transient state capacitance current

When analyzing transient state capacitance current, the r in Fig. 2 _ldo not consider with L, utilize L ₀, C, R ₀the series loop and the zero sequence sinusoidal voltage u that form ₀, according to equivalent circuit diagram, write out the differential equation of transient state capacitance current

In formula: the amplitude that Um is residual voltage.

Transient state capacitance current iC is by transient state free oscillation component i _c.oswith stable state power frequency component i _c.stdimerous, in fault moment starting condition, be i _c.os+ i _c.st=0, know again I _cm=U _mω C, tries to achieve transient state capacitance current and is

In formula:

I _cmamplitude for transient state capacitance current; ω _fangular frequency for transient state free oscillation component; δ=1/ τ _c=R/2L ₀for the attenuation coefficient of free oscillation component, the time constant that τ C is wherein loop; The phase angle of phase voltage when φ is fault.

When time, the transient state process of loop current has periodic vibration and attenuation characteristic; When time, loop current has aperiodic oscillatory extinction characteristic, and the state that tends towards stability gradually.

2. transient state inductive current

The inductive current of arc suppression coil is comprised of the DC component of transient state and the AC compounent of stable state, is expressed as follows:

In formula:

τ _ltime constant for inductor loop; ; The phase angle of phase voltage when φ is fault.

The oscillation amplitude of transient state process is relevant with the phase angle φ of ground connection moment supply voltage, and when φ=0, it is worth maximum; When φ=pi/2, it is worth minimum.

For isolated neutral and two kinds of modes of neutral by arc extinction coil grounding, the amplitude of the transient current at the initial stage of breaking down and frequency all mainly determined by transient state capacitance current, and phase angle is relevant simultaneously and initially for its amplitude.

In the time of near fault occurs in phase voltage instantaneous value maximal value, the transient state component of capacitance current has maximal value; When singlephase earth fault occurs in phase voltage instantaneous value, be near zero time, the transient state component of capacitance current is very little.

While there is singlephase earth fault in oil-filled power transformer peripheral circuit, known by analyzing transient state process, transient signal exists abundant failure message, and transient state process is not subject to the impact of earthing mode, therefore, study relevant signal processing method and analyze transient state component, the failure message comprising in research transient state component, has very important meaning for distribution network On-line Fault Detection and diagnosis.

Due to the long-time running of oil-filled power transformer, the insulating material that it is inner, under heat and electric effect, aging and decomposition gradually, and slowly produce the gases such as a small amount of various low molecular hydrocarbons and carbon monoxide, carbon dioxide.Under normal circumstances, decomposition is very slowly.But when there is latency fault, generation speed and the concentration of these gases will increase.

Along with further developing of fault, the gas decomposing out will constantly be dissolved in oil.Empirical evidence, H in transformer oil ₂, CH ₄, C ₂h ₆, C ₂h ₂, CO, CO ₂deng the component concentration of characteristic gas and the type of fault and the order of severity, there is substantial connection.Gas composition feature is also along with the difference of fault type, fault energy and the insulating material that relates to thereof and difference.

When break down in power transformer inside, between its phenomenon of the failure, failure cause and trouble location, there is a large amount of uncertainties.Being related to of fault type and coding itself is fuzzy, and often a kind of malfunction may cause various faults feature, and a kind of fault signature also can reflect various faults state in varying degrees.Thereby, in fail-safe control, by traditional accurate mathematical theory, be difficult to describe relation therebetween, very difficult true fault and the reason of diagnosing out transformer.In the division of the gas section of IEC three-ratio method, separation 0.1,1 and 3 be according to a large amount of transformer examples through investigation, analyze, adopt mathematical statistics method to draw.Obviously, the value of critical point is a statistical value, has certain dispersiveness, and this statistical method has inevitably been given up some secondary causes, must take that to sacrifice certain accuracy of judgement degree be cost.

IEC three-ratio method is selected 5 kinds of characteristic gas C exactly ₂h ₄, C ₂h ₆, C ₂h ₂, CH ₄, H ₂form three correlative value C ₂h ₂/ C ₂h ₄, CH ₄/ H ₂and C ₂h ₄/ C ₂h ₆, identical in the situation that these ratios with different coded representations, as shown in table 1.According to test result calculations, draw coding, and three correlative values are converted into corresponding code set, the correspondence of then tabling look-up draws the large body region of fault type and fault, as shown in table 2.

Table 1 coding rule

Table 2 fault type determination methods

Membership function in fuzzy mathematics can be described this dispersed phenomenon on mathematical statistics well.

C ₂h ₂/ C ₂h ₄the structure of ratio coding subordinate function

While processing encoded question by fuzzy mathematics, the fuzzy interval that correct structure coding membership function is portrayed frontier point, is the foundation stone that fuzzy set is applied to dissolved gas analysis, is also the key of faults strictly according to the facts.The deterministic process of membership function, is objectively in essence, but allows certain subjectivity.Not general theorem or the computing formula of a unification of the foundation of subordinate function in fuzzy mathematics, conventionally rule of thumb or statistics determine, definite method of commonly using has method of subjective appraisal, fuzzy statistical method, binary to contrast ranking method etc.

If C ₂h ₂/ C ₂h ₄=x, works as C ₂h ₂/ C ₂h ₄coding be respectively 0,1,2 o'clock, its mathematical function model characteristic function of a set C _r(X) be expressed as follows:

$C_0\left(x\right)=\left\{\begin{array}{cc}1,& x<0.1\\ 0,& x\&GreaterEqual;0.1\end{array}\right.---\left(2.1\right)$

Wherein R presentation code 0,1,2, when x≤1, and C ₀(x)=1 represents that ratio now belongs to the set that is encoded to 0 utterly; When x>=1, C ₀(x)=0 represents that ratio now does not belong to the set that is encoded to 0 utterly.Be encoded to 0 and 1 characteristic function of a set C ₀(x) figure as shown in Figure 9 and Figure 10.

From Fig. 9 and Figure 10, can find out, at separation 0.1 place, this ratio belongs to coding 1, and is only less than 0.1,

Coding is 0 with regard to saltus step, has a saltus step at 0.1 place.And in fact ratio be that 0.1 o'clock coding should incorporate into is 0 or 1 to be difficult to certainly actually, maximum in 0.1 place's ambiguity.In actual fault example, find, sometimes should incorporate into as encoding 0.1 0 be correct, and sometimes 0.1, incorporating into is 1 to be also correct, and therefore at 0.1 place, coding both can be 0 and also can be 1.

But be certainly less than 0.1 but near 0.1 place, it is large than the degree that belongs to 1 that coding belongs to 0 degree, and be greater than 0.1 but near 0.1 place, encode that it is larger than the degree that belongs to 0 to belong to 1 degree, and be not fully 2 to belong to 0 or 1.

In order to describe this fuzzy phenomena, with the membership function in fuzzy mathematics, carry out Fuzzy processing, with membership function, replace original fundamental function, the variation by by 1 to 0, is converted into slowly by steepness.Under the situation of the corresponding figure of known original encoding characteristic function of a set, according to existing experience, with reference to Fuzzy Distribution common and important in fuzzy mathematics, select and fall the mathematical model that ridge shape distributes to describe this Interval Fuzzy.Falling ridge shape distributes as shown in figure 11.

Parameter a wherein ₁, a ₂about an x=(a _l+ a ₂)/2 symmetry, at separation x=(a ₁+ a ₂)/2 u of place (x)=0.5, at a ₁, a ₂between be a kind of level and smooth transition.Parameter a ₁, a ₂value by concrete condition, determined.

With regard to C ₂h ₂/ C ₂h ₄fundamental function C ₀(x), parameter a in Figure 11 ₁, a ₂need satisfied condition:

$\left\{\begin{array}{c}(a_1+a_2)/2=0.1\\ 0<a_1<0.1\\ a_2>0.1\\ 0.1-a_1=a_2-0.1\end{array}\right.---\left(2.4\right)$

According to above-mentioned restrictive condition, in conjunction with concrete experience and statistical study, show a _lget 0.08, a2 and get 0.12, can portray more exactly this fuzzy interval, the distribution function that falls ridge shape distribution is shown below:

$\mathrm{\&mu;}\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;a_1\\ 0.5-0.5\sin \frac{\mathrm{\&pi;}}{a_2-a_1}(x-\frac{a_1+a_2}{2}),& a_1<x\&le;a_2\\ 0& x>a_2\end{array}\right.---\left(2.5\right)$

A ₁=0.08, a ₂=0.12 substitution above formula, can obtain actual Fuzzy Distribution function u ₀(x) be

$u_0\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.08\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-0.1)& 0.08<x\&le;0.12\\ 0,& 0.12<x\end{array}\right.---\left(2.6\right)$

In formula, x=C ₂h ₂/ C ₂h ₄, u ₀(x) described C ₂h ₂/ C ₂h ₄this ratio is under the jurisdiction of the degree of membership of 0 this coding.

Equally, for C ₂h ₂/ C ₂h ₄the subordinate function that is under the jurisdiction of coding 1, maximum at separation 0.1 and 3 place's ambiguities.Selected function must meet, u ₁(0.1)=0.5, u ₁(3)=0.5.In conjunction with known separation, Fuzzy Distribution selects ridge shape as shown in figure 12 to distribute.Parameter is consistent with aforesaid value, gets a ₁=0.08, a ₂=0.12.Should meet following requirement:

$\left\{\begin{array}{c}\frac{a_3+a_4}{2}=3\\ 0.1<a_3<3\\ a_4>3.0\\ 3.0-a_3=a_4-3.0\end{array}\right.---\left(2.7\right)$

Being chosen at of parameter meets not uniquely under above-mentioned condition to be determined, is convenient to again be calculated as when can portray fuzzy interval preferably.In conjunction with actual conditions, get a ₃=2.9, a ₄=3.10.

The function expression distributing with reference to ridge shape in Fuzzy Distribution, through definite parameter value, can obtain the expression formula of actual Fuzzy Distribution function suc as formula institute's formula, in conjunction with oneself

$u_1\left(x\right)=\left\{\begin{array}{cc}0,& x<0.08\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 1,& 0.12<x\&le;2.9\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-3.0)& 2.9<x\&le;3.1\\ 0,& 3.1<x\end{array}\right.---\left(2.8\right)$

U ₁(x) C has been described ₂h ₂/ C ₂h ₄this ratio is under the jurisdiction of the degree of 1 this coding.

Adopt parameter determination method same as described above, in conjunction with traditional experience separation, select and fall ridge shape and distribute to describe ratio C ₂h ₂/ C ₂h ₄the degree of membership that is under the jurisdiction of 2 these codings.Concrete membership function is:

$u_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;2.9\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-3),& 2.9<x\&le;3.1\\ 1,& 3.1<x\end{array}\right.---\left(2.9\right)$

Like this, Fuzzy Distribution function is combined with traditional experience cut off value, portrayed more accurately the fuzzy interval at separation place, fuzzy interval sharpening.

CH ₄/ H ₂the structure of ratio coding subordinate function

Select same Fuzzy Distribution, in conjunction with different separately separations, get corresponding parameter value, obtain CH ₄/ H ₂be under the jurisdiction of respectively the Fuzzy Distribution function of 0,1,2 codings

$v_0\left(x\right)=\left\{\begin{array}{cc}0,& x<0.08\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 1,& 0.12<x\&le;0.9\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-1.0),& 0.9<x\&le;1.1\\ 0,& 1.1<x\end{array}\right.$

$v_1\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.08\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 0,& 0.12<x\end{array}\right.$

$v_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;0.9\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-1)& 0.9<x\&le;1.1\\ 1,& 1.1<x\end{array}\right.$

C ₂h ₄/ H ₂the structure of ratio coding subordinate function

In like manner, C ₂h ₄/ C ₂h ₆the function w that is under the jurisdiction of respectively 0,1,2 codings ₀(x), w _l(x), w ₂(x),

$w_0\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.9\\ 0.5+0.5\sin 5\mathrm{\&pi;}(x-1),& 0.9<x\&le;1.1\\ 0,& 1.1<x\end{array}\right.---\left(2.13\right)$

$w_0\left(x\right)=\left\{\begin{array}{cc}0,& x<0.9\\ 0.5+0.5\sin 5\mathrm{\&pi;}(x-1),& 0.9<x\&le;1.1\\ 1,& 1.1<x\&le;2.9\\ 0.5-0.5\sin 5\mathrm{\&pi;}(x-.30),& 2.9<x\&le;3.0\\ 0,& 3.1<x\end{array}\right.---\left(2.14\right)$

$w_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;2.9\\ 0.5+0.5\sin 5\mathrm{\&pi;}(x-3.0),& 2.9<x\&le;1.1\\ 1,& 3.1<x\end{array}\right.---\left(2.15\right)$

For one group of concrete test figure, by constructing corresponding IEC tri-ratios, and utilize the Fuzzy Distribution function of three ratios, just can try to achieve them for the subjection degree of different coding.According to maximum membership degree method, just can determine even μ of corresponding fault type again _j=max (μ ₁, μ ₂,, ¨, μ _n), think that failure cause is j kind fault.Wherein, the parameter in distribution function is under certain condition restriction, and the figure while determining cut off value in conjunction with original statistical study is suitably chosen, the frequency of reflection statistics to greatest extent value.The value of parameter is not unique, further carrying out feedback adjusting in fuzzy analysis, finds best value.Selected Fuzzy Distribution sometimes needs to verify according to the accuracy of diagnosis in comprehensive judge later, can repeatedly revise, until comparatively conform to the fact.

IEC three-ratio method after Fuzzy processing can be described the coding characteristic of transformer test data more accurately.

In order to improve the accuracy of IEC three-ratio method, the present invention introduces Vague Diagnosis Technique, and IEC three-ratio method is carried out to Fuzzy processing, can describe more accurately the coding characteristic of transformer test data.

Adopt integrated method that visualization technique and Vague Diagnosis Technique are combined, effectively reduced that the inevitable diagnosis bringing while adopting single transformer fault control method to diagnose makes mistakes or the diagnostic result erroneous judgement problem such as comprehensively not.To a certain extent, increased the accuracy of diagnosis.

Artificial neural network is by a large amount of and the interconnected network forming of the similar artificial neuron of natural neurocyte.The mode that neural network is dealt with problems is completely different from traditional statistical method, it is the thinking of simulation human brain, a large amount of neurons is linked to be to a complicated network, utilize oneself to know that sample trains network, be similar to the study of human brain, allow nonlinear relationship between network storage variable, be similar to the memory function of human brain, then utilize the network information of storage unknown sample is classified or predict, be similar to the association function of human brain.General neural network is all adjustable, trainable in other words, and so specific input just can obtain the output requiring.

The key of transformer fault control technology is the mapping realizing from Fault symptom space to defective space, thereby realizes the identification of fault and diagnosis.Utilize detection data to carry out fail-safe control to filling transformer, be actually gas composition to a complicated Nonlinear Mapping of fault type.

If reflection transformer fault sign vector set is combined into X _p∈ rR ⁿ(P=1,2 ..., Q), it is the set of Q m dimensional vector, the set of corresponding fault mode vector is Y _p∈ R ^m, it is the set of Q m dimensional vector.If they form sample pair set Ω={ X _p, Y _pexpressed the internal relation of failure symptom and fault mode, and the existence mapping F corresponding with set omega meets:

F:X→YY=F(X)(3.4)

F has reflected the internal relation of failure symptom and fault mode, deserves to be called and states failure symptom vector to the solution that is mapped as fail-safe control problem of fault mode vector.

In addition, establishing F is that failure symptom vector is to the mapping of fault mode vector,

F:(R ⁿ) _p→(R ^m) _pY=F(X)(3.5)

There is mapping function M in embodiment of the present invention diagnostic model

M:(R ⁿ) _p→(R ^m) _pY=M(X)(3.6)

If any x ∈ X _pcorresponding Y=F (X _p) and Y=M (X) all meet:

‖Y-Y′‖≤ε(3.7)

‖ in formula. ‖ is the norm on sample space; ε is given permissible error, can think that mapping M and mapping F are approximately equalised under given error ε, i.e. M ≈ F.If mapping F is the solution of fail-safe control problem, M is called the approximate solution of fail-safe control problem, i.e. the output of this artificial neural network has reflected fault mode, and this model can be used as transformer fault and controls.

Conventionally mapping relations F is unknown.But neural network can be input to by study the sample set of output, realize this mapping relations that are input to output, its mechanism of action can represent with Fig. 3.

Because artificial neural network has good mode identificating ability, by the study to great amount of samples, can when thering is minimum computing, can set up the Nonlinear Mapping relation of phenomenon of the failure and reason, can store the pattern of arbitrary continuation to (X _k, T _k; K=1,2 ..., m), complete input { X _kto desired output { Z _kmapping.Therefore,, in transformer fault control technology of the present invention research, adopt BP network (being counterpropagation network, back propagation).

BP network reason ultimate principle is as Fig. 4.It is a typical feedforward hierarchical network, structurally has following features:

1. network is divided into input layer I, hidden layer H and output layer O.According to BP network theory, hidden layer can be one or more layers, but has oneself warp of BP network of a hidden layer with arbitrary accuracy, to complete the mapping of any continuous function, and the present invention is excessively slow for fear of training speed, determines only to adopt a hidden layer.

2. I layer has n node, the n of map network sign input (x ₁, x ₂..., x _n); O layer contains q node, with the q kind fault output (Z of BP network ₁, Z ₂..., Z _q) response.H node layer (y ₁, y ₂..., y _p) needs that come according to each diagnostic criteria of physical simulation and method of number arrange.

3. with node layer onrelevant, different node layer forward connection.

Making I node layer is v to the connection weight between H node layer _hi, V=V _nxp; Making H node layer is w to the connection weight between O node layer _ij, W=W _pxq.K (k=1,2 ..., m) be given sample number.θ _ifor the thresholding of H node layer, γ _ithresholding for O node layer.

Global error between the desired output of the network based sample of BP and actual output, by learning process, from output layer, layer-by-layer correction weight coefficient, makes both difference be less than the numerical value of regulation.

Through the BP network of learning training, after input fault sign vector, it exports T _kwith desired output Z _kunder permissible error, be approximately equalised.Therefore, the output of BP network is exactly the approximate of physical fault pattern, thereby obtains the paradiagnosis solution of failure problems.

BP network learning and training process is comprised of forward-propagating output procedure and backpropagation adjustment process two parts.In forward-propagating process, input signal is successively processed through hidden layer from input layer, and is transmitted to output layer, and the neuronic state of every one deck only affects the neuronic state of lower one deck; If can not get the output of expectation at output layer, the error of output signal is by the connecting path backpropagation along original, until input layer; By revising each layer of interneuronal connection weights on the way, make the global error between desired output and actual output progressively reach minimum.

To each pattern to (x ^k, T ^k) (k=l, 2 ..., m) by FB(flow block) 5, learn:

1. use random number initialization input layer I and hidden layer H, the weight coefficient V between hidden layer H and output layer O _hi, W _ijselected step-length η, 0< η <1; L ← 1, l is used for counting iterations, and maximum iterations is taken as L; M ← 1, m is used for counting sample number; Permissible error is taken as Emax.

2. by input pattern X _kdeliver to I layer, the activation value X of I layer unit _hby connection weight v matrix, deliver to H layer, produce the new clean input nety of H layer, and then produce the output valve y of H layer unit _i,

${\mathrm{nety}}_i=\underset{h-1}{\overset{n}{\mathrm{\&Sigma;}}}{v}_{\mathrm{hi}}{x}_h+{\mathrm{\&theta;}}_i---\left(3.8\right)$

y ₁=f(nety _j)(3.9)

In formula, i=1,2 ..., p;

Because Sigmoid activation function has the non-linear of height, and there is good expressive ability while using together with BP learning algorithm, therefore excitation function f adopts monopole type s type function.

3. calculate the output valve of O layer unit:

$z_j=f(\underset{i=1}{\overset{p}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{y}_i+r_j)---\left(3.10\right)$

In formula, j=1,2 ..., q.

4. calculate the vague generalization error of output layer Unit 0:

$d_j=z_j(1-z_j)(T_j^k-z_j)---\left(3.11\right)$

In formula, j=1,2 ..., q; K=1,2 ..., m (m is total sample number); for the desired output of O layer unit j, it is teacher signal

5. calculate the vague generalization error of output H layer unit:

$e_i=y_i(1-y_i)\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_{\mathrm{ij}}{d}_j---\left(3.12\right)$

In formula, i=1,2 ..., p; Above formula is equivalent to the error back propagation of O layer unit to H layer.

6. adjust.Layer unit arrives connection weight and the thresholding of H layer unit:

Δw _ij=ηy _id _j’Δr _j=ηd _j(313)

In formula, i=1,2 ..., p; J=1,2 ... q, η; η is learning rate (0< η <l).

7. adjust connection weight and thresholding that H layer unit arrives I layer unit:

Δv _hi=ηx _ke _i，Δθ _i=ηe _i(3.14)

In formula, h=l, 2 ..., n; I=1,2 ..., q.

8. renewal learning pattern, if p<m, p ← p+1, turns the 2. step; Otherwise mode of learning finishes to turn the 9. step.

9. renewal learning number of times.l←l+1。If l>L, iteration finishes, otherwise:

If e<Emax, iteration finishes;

If.E >=Emax.E ← 0, p ← 1, turns the 2. step, enters next round iteration.

The I/O mode-definition of each diagnostic criteria is as follows:

1. improve Rogers three-ratio method: the input pattern of this diagnostic criteria is a vector being comprised of following three ratios.

[x]=[CH ₄/H ₂,C' ₂H ₂/C ₂H _q,C ₃H _q/C _ZH ₆] ^r

Output mode is six type vectors of determining according to transformer situation, as shown in table 3.For example transformer normal condition is by first element is 1, all the other elements are 0 vector representation.

[y]=[l，0，0，0，0，0，] ^「

Neural network after finishing for training, then adopt the reliability of training sample set pair neural network to identify checking.

Its recognition result is as shown in table 3.

Table 3 uses the BP Network Recognition result of improving rogers standard.

2. IEC three-ratio method: the input pattern of this diagnostic criteria is a vector being comprised of following three ratios.

[x]=[cH ₄/H ₂,c _tH ₂/c ₂H ₄,C _,H ₄/C ₂H ₅] ^r

Output mode is six type vectors of determining according to transformer operation conditions, as shown in table 4.For example transformer fault caused by low energy discharge is by second element is 1, all the other elements are 0 vector representation.

[y]=[0，1，0，0，0，0] ^T

Neural network after finishing for training, then adopt the reliability of training sample set pair neural network to identify checking.

Its recognition result is as shown in table 4.

Table 4 uses the BP Network Recognition result of IEC standard.

While carrying out neural metwork training, improve the variation tendency of Rogers standard population error and see Fig. 6.When hidden layer is 18 neurons, algorithm reaches global error 0.14365 after 60000 iteration of adjustment weights.In the present embodiment, this global error is acceptable, selects as neural network test, and its test result is shown in Fig. 8.

The neural metwork training of IEC standard the results are shown in Figure 7.When hidden layer is 16 neurons, algorithm reaches global error 0.15198 after 30000 iteration of adjustment weights, and this global error is acceptable in the present embodiment, selects as neural network test, and its test result is shown in Fig. 8.

From the test result of the aforementioned neural network performance that each is shown, use the neural network of two kinds of distinct methods all to there is higher diagnosis efficiency, this is due to known these data of neural network.Mutual relationship between the I/O variable of neural network, quantity and each variable, has a certain impact to the effect that realizes of neural network.In general, if neural metwork training speed is slow or be difficult to convergence, just show that this I/O pattern dependency is poor or incorrect.From Fig. 6 to Fig. 7 neural metwork training process, after certain iterations, neural network can both restrain.But the diagnostic criteria of using or method are different, the speed of its speed of convergence is also just different.This is because the mutual relationship between I/O variable, quantity and each variable is to be determined by different diagnostic criteria or method.

From Fig. 6 to Fig. 7 neural metwork training process, the diagnostic criteria of using or method are different in addition, and the impact that hidden nodes is realized effect to neural network is not identical yet.At present, choosing of hidden nodes k value also do not have reliable theoretical foundation.K value is chosen the too training of macroreticular will become difficult, and easily occur the redundancy phenomenon of hidden nodes; K value is chosen too little, and neural network is difficult to again guarantee that transformer fault controls due precision.Consider problem to be solved and desired computational accuracy, after geometric mean rule, the present invention adopts twice I/O neuron sum to carry out guestimate.Then in the learning process of neural network, the structure of self-organization and self study oneself.By continuous adjustment hidden nodes, obtain a best network structure.As the three-layer neural network of design has m input block and n output unit, hidden nodes is adjusted near 2 (m+n).

Claims (10)

1. transformer and a peripheral circuit fault control system thereof, is characterized in that, comprising: signal detecting part, feature extraction parts, state recognition parts and forecast decision component.

2. transformer and peripheral circuit fault control system thereof according to claim 1, is characterized in that, described signal detecting part, for measuring the status signal relevant with fault; Its signal detection process comprises following aspect: 1) signal measures process: the sniffer consisting of electric weight or non-electrical sensor obtains the transformation of variable of measuring object ginseng; 2) signal condition process: the signal that sniffer is obtained converts and transmits; 3) data acquisition: signal from becoming continuously discrete process; According to different diagnostic purposes and object, select can reflect the signal of measurand state.

3. transformer and peripheral circuit fault control system thereof according to claim 1, it is characterized in that, described feature extraction parts, for the signal conversion detecting is carried out to feature extraction, the initial state signal being obtained by input is passed through to signal analysis and processing, the feature of extracting energy faults state, forms pattern to be checked.

4. transformer and peripheral circuit fault control system thereof according to claim 1, it is characterized in that, described state recognition parts, for analysis of failure case, utilize database technology to set up fault file storehouse, the state model in fault file storehouse is decided to be to reference mode, calling pattern to be checked by the state model obtaining after input, feature extraction after kainogenesis fault, and compares judgement with reference mode.

5. transformer and peripheral circuit fault control system thereof according to claim 1, is characterized in that, described forecast decision component, for through differentiating, detects belonging to the continuation of normal condition, repeats above operation; To belonging to abnormality, find out failure condition, estimate Future Development and the time of continuing operation, and propose control measure and maintenance decision according to failure cause.

6. transformer and peripheral circuit fault control system thereof according to claim 1, is characterized in that, described signal detecting part further comprises transient state equivalent circuit; Described transient state equivalent circuit, is further used for analyzing transient state capacitance current; Described transient state equivalent circuit when analyzing transient state capacitance current, r _ldo not consider with L, utilize L ₀, C, R ₀the series loop and the zero sequence sinusoidal voltage u that form ₀, according to following differential equation, calculate transient state capacitance current:

In formula: U _mamplitude for residual voltage;

C represents the zero sequence equivalent capacity of electrical network;

L ₀for the zero sequence equivalent inductance containing transformer and line inductance;

R ₀represent the substitutional resistance in zero sequence loop;

R _l, L represents respectively active loss resistance and the inductance of arc suppression coil; u ₀for residual voltage.

7. transformer and peripheral circuit fault control system thereof according to claim 1, is characterized in that, utilizes 5 kinds of characteristic gas C ₂h ₄, C ₂h ₆, C ₂h ₂, CH ₄, H ₂form three correlative value C ₂h ₂/ C ₂h ₄, CH ₄/ H ₂and C ₂h ₄/ C ₂h ₆, identical in the situation that these ratios with different coded representations, as shown in table 1 below, according to test result calculations, draw coding, and three correlative values be converted into corresponding code set, the correspondence of then tabling look-up draws fault type and trouble location.

8. according to transformer described in claim 7 and peripheral circuit fault control system thereof, its feature exists, and described coding rule is as follows:

9. according to transformer described in claim 7 and peripheral circuit fault control system thereof, its feature exists, and described fault type judgment rule is as follows:

。

10. a transformer fault control system, is characterized in that, comprising: power transformer peripheral circuit detection subsystem, and, the inner detection subsystem of power transformer;

In the inner detection subsystem of described power transformer, the make of CH4/H2 ratio coding subordinate function is: CH ₄/ H ₂the Fuzzy Distribution function that is under the jurisdiction of respectively 0,1,2 codings:

$v_0\left(x\right)=\left\{\begin{array}{cc}0,& x<0.08\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 1,& 0.12<x\&le;0.9\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-1.0),& 0.9<x\&le;1.1\\ 0,& 1.1<x\end{array}\right.$

$v_1\left(x\right)=\left\{\begin{array}{cc}1,& x\&le;0.08\\ 0.5-0.5\sin 25\mathrm{\&pi;}(x-0.1),& 0.08<x\&le;0.12\\ 0,& 0.12<x\end{array}\right.$

$v_2\left(x\right)=\left\{\begin{array}{cc}0,& x\&le;0.9\\ 0.5+0.5\sin 25\mathrm{\&pi;}(x-1)& 0.9<x\&le;1.1\\ 1,& 1.1<x\end{array}\right.$

The inner detection subsystem of described power transformer, further comprises neural network failure control assembly; The inner detection subsystem of described power transformer is input to the sample set of output by described neural network learning, realization is input to the mapping relations of output, the Nonlinear Mapping relation of phenomenon of the failure and reason can be when thering is minimum computing, can set up, the pattern of arbitrary continuation can be stored to CX _k, T _k; K=1,2 ..., m), complete input { X _kfly to desired output { Z _kmapping.