CN102567783A - Expert fault analytical and diagnostic method of parallel mixed type power quality regulator - Google Patents

Abstract

The invention discloses an expert fault analytical and diagnostic method of a parallel mixed type power quality regulator. The method comprises the following steps that: a voltage signal is collected at a sampling point of a power quality regulator and is used as a fault sample signal; denoising processing is carried out on the collected fault sample signal; wavelet packet transform is utilized to carry out processing on the denoised fault sample signal so as to extract an energy characteristic vector; the extracted energy characteristic vector used as a target input vector, together with a set fault type output vector, is input into a BP neural network to carry out training, so that a fault classifier is established; and fault signals that are collected in real time are input into the fault classifier that compares the fault signals with various fault characteristics in the fault classifier, so that a fault disgnosis result is output. According to the invention, a fault state can be accurately diagnosed, so that losses caused by faults occur on a core component in the power quality regulator can be avoided or reduced.

Description

Expert's fault analysis of parallel connection mixed type electric energy regulator and diagnostic method

Technical field

The present invention relates to a kind of circuit failure diagnosis method, particularly a kind of parallel connection mixed type electric energy regulator expert's fault analysis and diagnostic method.

Background technology

The electric energy regulator of using in commercial production and the daily life can produce a large amount of harmonic waves, and is serious to electric network pollution.Because electric energy regulator carries out comprehensive compensation to the quality of power supply of electrical network in the concentrated area, can solve the power quality problem of electrical network comprehensively, become a research focus in recent years.Wherein most of research concentrates on the electric energy regulator main circuit topological structure, aspects such as harmonic current detecting method and control method, still, for the faults analysis of electric energy regulator self and Studies on Diagnosis but seldom.Yet; In a single day electric energy regulator breaks down; The consequence that its fault caused is very serious, and the actual motion of 35kv parallel connection mixed type electric energy regulator shows that most of faults show as the damage of device for power switching and interlock circuit thereof; Such as IGBT, the damage of these core components of thyristor and interlock circuit.

After these core components break down, fault is carried out quick diagnosis can reduce breakdown loss, therefore, necessary fault analysis and diagnosis to electric energy regulator conducts a research.

The major failure relevant with IGBT and circuit thereof has:

(1) bridge arm direct pass fault.When an IGBT on a certain brachium pontis begins conducting, and another can not in time turn-off because of breaking down, or when an IGBT during in conducting and another because of the false triggering conducting, the phenomenon of shoot through then appears on this brachium pontis.

(2) driving circuit fault.Because the influence of factors such as the interference of space electromagnetism and ground wire and environment may cause pulse producer output pulse disorderly, injures the normal operation of IGBT.

(3) superpotential fault.IGBT is conducting in the turn off process, because of the rapid variation of main circuit current, causes the major loop wiring inductance to induce high voltage, and very little circuit inductance just possibly cause very big Ldi/dt, thereby is easy to cause the superpotential that jeopardizes IGBT safety.

(4) overcurrent fault.Because IGBT is operated under the environment of high voltage, big electric current, and its volume is little, thermal capacity is little, when the electric current of overrate flow through, heat had little time to distribute, and caused IGBT to burn.Overcurrent fault takes place nothing more than two kinds of situation are arranged in IGBT; A kind of is exactly the fault of electric energy regulator self; As installing short circuit overcurrent that the aging short circuit overcurrent that causes of major component, IGBT brachium pontis short circuit that the direct current capacitors short circuit causes and bridge arm direct pass fault cause or the like, another kind is exactly the IGBT overcurrent that the fault that takes place of electric system or disturbance cause.

(5) du/dt fault.Be in the IGBT under the off state, because the rejuvenation of diode antiparallel with it, with the rapid rising of bearing voltage between C-E.This static du/dt leads at IGBT and produces the current direction gate driver circuit in the electric capacity at collection, between emitter-base bandgap grading, makes V under the worst situation _GEIncrease and (trend towards V _GEAnd reach threshold voltage (th)), cause IGBT to be switched on, the shoot through fault occurs.

(6) short trouble.In 35kv parallel connection mixed type electric energy regulator; When inverter circuit generation load short circuits that main switch element IGBT constitutes or same bridge arm direct pass fault; Busbar voltage directly is added in C, the E end of IGBT; The collector current that flows through IGBT can sharply increase, this moment as the untimely gate drive signal of removing, and IGBT will burn.

(7) overheating fault.Because characteristic and the temperature of the PN junction of IGBT are closely related, each IGBT working temperature can not be above its high workload junction temperature T _jDevice surpasses the maximal value of junction temperature and power when work, the numerous characteristics of device and parameter all can change, and the permanent phenomenon of burning of device occurs.IGBT is operated under the condition of high voltage, big electric current, and in switching process, surge current, inverse current have reached factors such as superpotential, all can cause the loss of IGBT switch power excessive, cause that entire chip is overheated.

The fault relevant with thyristor mainly contains:

(1) thyristor element fault.The thyristor element makes PN junction breakdown (short circuit) because of overtension, or crosses ambassador's PN junction because of electric current and blown (open circuit).

(2) the non-all-wave conducting of thyristor.Reactive power compensator will make the non-all-wave conducting of thyristor because the load of circuit is a power capacitor if current signal makes a mistake.

(3) thyristor phase shortage conducting.Bigger because of being parallel to the capacitance discharges resistance value, the voltage attenuation time is longer, might occur that certain situation of organizing not conducting is the conducting of thyristor phase shortage in 3 groups of thyristor valves.

Therefore, 35kV parallel connection type mixed type electric energy quality regulator is carried out above-mentioned fault analysis and diagnosis research has great importance, can reduce because the stop time that equipment failure causes; Reduce failure rate; Improve reliability, the accident that prevents takes place, and reduces maintenance cost.

Summary of the invention

In order to solve the above-mentioned technical matters that parallel connection type mixed type electric energy quality regulator fault diagnosis exists, the present invention provides a kind of parallel connection mixed type electric energy regulator expert's fault analysis and diagnostic method.

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

(1) from the sampled point acquired signal of electric energy regulator as the fault sample signal;

(2) the fault sample signal of gathering is carried out denoising;

(3) vector of the fault sample signal extraction energy feature after utilizing wavelet package transforms to denoising;

(4) with the energy feature vector that extracts as the target input vector, and be input to the BP neural network together with the fault type output vector of setting and train, set up fault grader;

The energy feature vector input fault sorter input fault sorter of the fault-signal that (5) will gather in real time, fault grader compares all kinds of fault signatures in fault-signal and the fault grader, the output fault diagnosis result.

Above-mentioned expert's fault analysis of parallel connection mixed type electric energy regulator and diagnostic method, the concrete steps of the fault sample signal of gathering being carried out denoising are:

(1) the fault sample signal f (k) that gathers is done wavelet transformation, obtain one group of wavelet coefficient w _{J, k}J, k representes scale factor, is integer;

(2) pass through w _{J, k}Soft, hard threshold values function carries out threshold values to be handled, and obtains estimating wavelet coefficient

Make

Minimum;

(3) utilize

and carry out wavelet reconstruction, obtain the signal after estimated signal

is denoising.

Above-mentioned expert's fault analysis of parallel connection mixed type electric energy regulator and diagnostic method, described hard threshold values function does

$\widehat{w_{j,k}}=\left\{\begin{array}{cc}{w}_{j,k}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Soft threshold values function does $\widehat{w_{j,k}}=\left\{\begin{array}{cc}\mathrm{Sgn}\left(w_{j,k}\right)\left(\right|w_{j,k}|-\mathrm{\&lambda;}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Wherein sgn () is a sign function, and threshold values λ is

In above-mentioned the expert's fault analysis of parallel connection mixed type electric energy regulator and diagnostic method, the concrete steps of said fault sample signal extraction energy feature vector after utilizing wavelet package transforms to denoising are:

(1) the fault sample signal after the denoising is carried out n layer WAVELET PACKET DECOMPOSITION, extract 2 of n layer ⁿIndividual each band signal characteristic;

(2) each coefficient of dissociation of wavelet packet is carried out reconstruct, in band limits separately, form the signal S after the reconstruct;

(3) ask the gross energy of each band signal;

(4) tectonic energy measure feature vector.

Technique effect of the present invention is: the present invention carries out denoising to the voltage or the current failure sample of the electric energy regulator collection point of collection; Fault-signal after utilizing wavelet package transforms to denoising then carries out energy feature and extracts; Again the energy feature vector that extracts also being input to the BP neural network with the fault type output vector of setting together as the target input vector trains; The fault sample that to gather is in real time at last called in fault grader; And the fault grader comparison with automatic memory all kinds of fault signatures good with foundation; Can diagnose out corresponding malfunction, be convenient to the measure of in time gathering, thereby avoid and reduce the loss that causes when core component breaks down in the electric energy regulator.

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

Description of drawings

Fig. 1 is a Troubleshooting Flowchart of the present invention.

Fig. 2 is neural network algorithm design cycle among the present invention.

Fig. 3 is the main loop circuit structural drawing of parallel connection type mixed type electric energy quality regulator among the present invention.

Embodiment

Referring to Fig. 1, Fig. 1 is for being Troubleshooting Flowchart of the present invention.Carry out fault diagnosis, will gather curtage fault sample signal from electric energy regulator circuit sampling point earlier.Electric energy regulator circuit sampling point is chosen and is adopted the network method of Tearing, and the basic thought of method of Tearing is through " tearing " means, and a big circuit is resolved into some little circuit, and makes mutual independence between these little circuit, but parallel parsing calculates.The mixing that method of Tearing has node to tear, branch road is torn or both combine is torn.

The key of carrying out module level fault quick diagnosis for large-scale circuit is the circuit part node is carried out effectively repeatedly tearing accurately; Follow tearing principle and intersecting the criterion of tearing of tearing the search diagnosis of network method of Tearing when repeatedly tearing, specific as follows:

The numbering of (1) tearing node is organized at last;

(2) node serial number that belongs to same electronic circuit is (except that tearing node) continuously;

(3) have only and to reach node and just can tear, comprise common node;

(4) there is not the coupling between topological relation and the parameter between each child network after tearing;

When (5) tearing, the scale that as far as possible makes sub-network is for minimum, promptly with can reach node and can continue to tear again the time, should continue to tear.

When (6) tearing, keep the relative integrality of each sub-network structure as far as possible, should not tear like the symmetry tubes of differential amplifier, and should they be put in the same sub-network.

Method denoising with the fault sample signals collecting small echo floating threshold of gathering.In the real-time observation process of electric energy regulator, signal to be detected often is accompanied by a large amount of noises.If these signals are used as the fault diagnosis pilot signal, will possibly cause wrong diagnosis or fail to pinpoint a disease in diagnosis disconnected.Therefore, be necessary the noise in the detection signal is eliminated.

The fault sample signal that gather the collection point and the small echo denoising process of fault diagnosis pilot signal are following:

If the signal of gathering is f (t)=s (t)+n (t), wherein, s (t) is an original signal, and n (t) is that variance is σ ²White Gaussian noise, obey N (0, σ ²), N is the number of discrete signal.

We at first carry out discrete sampling to f (t), obtain N point discrete signal f (n), n=0, and 1,2 ..., N-1, its wavelet transformation does

$\mathrm{Wf}(j,k)=2^{-\frac{1}{2}j}\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}f\left(n\right)\mathrm{\&psi;}(2^{-j}-k)$ J, k representes scale factor, is integer;

Wf (j, k)) is wavelet coefficient.In practical application, the calculating of following formula is loaded down with trivial details, and wavelet function ψ (t) generally do not have explicit expression, thereby the Recursive Implementation method of wavelet transformation is arranged:

Sf(j+1，k)＝Sf(j，k)*h(j，k)

Wf(j+1，k)＝Sf(j，k)*g(j，k)

Wherein, H and g are respectively scaling function

and corresponding low pass and the Hi-pass filter of wavelet function ψ (t), and Sf (0, k) be original signal; Sf (j; K) be scaling function, (j k) is wavelet coefficient to Wf.Corresponding reconstruction formula does

$\mathrm{Sf}(j-1,k)=\mathrm{Sf}(j,k)*\stackrel{\&CenterDot;}{\stackrel{\&CenterDot;}{h}}(j,k)+\mathrm{Wf}(j,k)*\stackrel{\&CenterDot;}{\stackrel{\&CenterDot;}{g}}(j,k)$

Wherein,

and

corresponds respectively to reconstruct low pass and Hi-pass filter.

For simplicity, we remember w _{J, k}=Wf (j, k), j, k representes scale factor.Because wavelet transformation is linear transformation, so to after f (k)=s (k)+n (k) makes wavelet transform, the wavelet coefficient w that obtains _{J, k}Still be made up of two parts, a part is that (j k), is designated as u to the corresponding wavelet coefficient Ws of signal s (k) _{J, k}, another part is that (j k), is designated as v to the corresponding wavelet coefficient Wn of noise n (k) _{J, k}

The basic thought of small echo threshold values denoising method is: work as w _{J, k}During less than a certain critical threshold values, think at this moment w _{J, k}Mainly cause, give up by noise; Work as w _{J, k}During greater than this critical threshold values, think that wavelet coefficient at this moment mainly has signal to cause, so just this a part of w _{J, k}Directly remain (hard threshold values method) or press a certain fixed amount, carry out the signal after wavelet reconstruction obtains denoising by new wavelet coefficient then to zero contraction (soft threshold method).The method can realize through following three steps:

(1) signals with noise f (k) is done wavelet transformation, obtain one group of wavelet coefficient w _{J, k}

(2) pass through w _{J, k}Soft, hard threshold values function carries out threshold values to be handled, and obtains estimating wavelet coefficient Make

As far as possible little;

The hard threshold values function that uses does

$\widehat{w_{j,k}}=\left\{\begin{array}{cc}{w}_{j,k}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Soft threshold values function does $\widehat{w_{j,k}}=\left\{\begin{array}{cc}\mathrm{Sgn}\left(w_{j,k}\right)\left(\right|w_{j,k}|-\mathrm{\&lambda;}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Wherein sgn () is a sign function, threshold values λ go for N be counting of discrete signal.

(3) utilize

and carry out wavelet reconstruction; Obtain estimated signal ), be the signal after the denoising.

With the vector of the signal extraction energy feature after the denoising, its process is following:

(1) at first original signal is carried out n layer WAVELET PACKET DECOMPOSITION, extract 2 of n layer ⁿIndividual each band signal characteristic (low frequency coefficient and high frequency coefficient X _Ij).

(2) each coefficient of dissociation of wavelet packet is carried out reconstruct, in band limits separately, form the signal S after the reconstruct.

(3) ask the gross energy of each band signal.

If S representes original signal, d representes WAVELET PACKET DECOMPOSITION coefficient, S _IjBe reconstruction signal, S _IjCorresponding gross energy is E _{I, j}, d _{I, k}Expression S _IjThe amplitude (being the wavelet package reconstruction coefficient) of discrete point.Then have:

$E_{i,j}={\left|{\&Integral;S}_{i,j}\left(t\right)\right|}^2,\mathrm{dt}=\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left|d_{j,k}\right|}^2$

The number of plies i=0 that wherein decomposes, 1,2 ... N; J=0,1,2 ... 2 ⁿK=1,2 ... N; (i, j) j node of expression i layer.

Because WAVELET PACKET DECOMPOSITION process energy conservation, promptly every layer of energy all equates and equals gross energy, and then whole energy of definition signal are:

The relative wavelet-packet energy that defines certain frequency range is:

Defining relative wavelet-packet energy proper vector is: W _i=(M _0,0, M _0,1K K M _{I, j})

Choosing each frequency band relative energy formation energy feature vector of n layer is:

One of actual common method that belongs to pattern-recognition of neural network classifier, the energy feature that accurately extracts power electronic equipment circuit major loop fault is vectorial, just can confirm the mapping of fault-signal to the energy feature vector.Reach the mapping from the fault-signal to the fault type, also must confirm of the mapping of energy feature vector to running status.The process of setting up of these mapping relations from the energy feature vector to fault type is exactly the mode identification procedure of fault diagnosis, and its essence is a classification problem.Wherein, the most basic problem is exactly the definite and training process of the type of sorter, to establish the relation of energy feature vector sum running status.

The foundation of neural network failure sorter:

(1) sets input vector: with the voltage that extracts and current energy proper vector sample input sample as neural metwork training.

(2) set the export target vector: make for each failure condition, all have different output vectors corresponding.

(3) neural network type design: adopt the recurrent neural network that has deviation unit.

(4) network number of plies design: be designed to three layers.

(5) network algorithm design: as shown in Figure 2.

Intelligent trouble diagnosis

The denoising of primary fault sample, feature extraction and fault grader three partial functions of classifying are combined, just be built into off-line type intelligent trouble diagnosis system.When carrying out fault diagnosis; Energy feature vector input fault diagnostic system with signal to be tested; Contrast with object vector then; If a certain in the object vector matches with primary fault sample to be tested, then the diagnostic result of fault sample is the pairing fault type of this object vector, provides fault diagnosis result then.

At first be to confirm input vector.The sample signal energy feature vector that will after the wavelet-packet energy method is extracted, obtain is as the input sample of neural metwork training.Also need confirm output vector, make, all have different output vectors corresponding for each failure condition.The classification of parallel connection type mixed type electric energy quality regulator major loop fault type is as shown in table 1, establishes output vector Y=X ₁X ₂X ₃X ₄X ₅X ₆X ₇X ₈, the device that the Gao Siwei representative in the vector is broken down is represented fault type for low four, and device number is as shown in Figure 3.

The failure modes of table 1 parallel connection type mixed type electric energy quality regulator major loop

Failure modes	Fault type (object vector)
		The first kind (0000)	Non-fault
Second type (0001)	The bridge arm direct pass fault
		The 3rd type (0010)	The driving circuit fault
The 4th type (0011)	The superpotential fault
		The 5th type (0100)	Overcurrent fault

The 6th type (0101)	The du/dt fault
		The 7th type (0110)	Short trouble
The 8th type (0111)	Overheating fault
		The 9th type (1000)	The thyristor element fault
The tenth type (1001)	The non-all-wave conducting of thyristor
		The 11 type (1010)	The conducting of thyristor phase shortage

In actual experiment, choose 120 test sample books and carried out confirmatory experiment; Wherein there are 115 sample diagnostic results entirely true; Only there are 5 samples wrong diagnosis to occur; Table 2 is for enumerating out the fault diagnosis result of 8 fault samples to be tested, and fault type of diagnosing out and real system failure condition are basic identical, show that this diagnostic system has very high fault diagnosis rate.

Table 2 fault diagnosis test result

Claims (4)

1. parallel connection mixed type electric energy regulator expert's fault analysis and diagnostic method may further comprise the steps:

(1) from the sampled point acquired signal of electric energy regulator as the fault sample signal;

(2) the fault sample signal of gathering is carried out denoising;

(3) vector of the fault sample signal extraction energy feature after utilizing wavelet package transforms to denoising;

(4) with the energy feature vector that extracts as the target input vector, and be input to the BP neural network together with the fault type output vector of setting and train, set up fault grader;

The energy feature vector input fault sorter of the fault-signal that (5) will gather in real time, fault grader compares all kinds of fault signatures in fault-signal and the fault grader, the output fault diagnosis result.

2. parallel connection mixed type electric energy regulator expert's fault analysis according to claim 1 and diagnostic method, the concrete steps of said step (2) are:

(1) the fault sample signal f (k) that gathers is done wavelet transformation, obtain one group of wavelet coefficient w _{J, k}, j, k representes scale factor, is integer;

(2) pass through w _{J, k}Soft, hard threshold values function carries out threshold values to be handled, and obtains estimating wavelet coefficient

Make

Minimum;

(3) utilize

and carry out wavelet reconstruction, obtain the signal after estimated signal

is denoising.

3. parallel connection mixed type electric energy regulator expert's fault analysis according to claim 2 and diagnostic method, described hard threshold values function does

$\widehat{w_{j,k}}=\left\{\begin{array}{cc}{w}_{j,k}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Soft threshold values function does $\widehat{w_{j,k}}=\left\{\begin{array}{cc}\mathrm{Sgn}\left(w_{j,k}\right)\left(\right|w_{j,k}|-\mathrm{\&lambda;}& \left|w_{j,k}\right|\&GreaterEqual;\mathrm{\&lambda;}\\ 0& \left|w_{j,k}\right|<\mathrm{\&lambda;}\end{array}\right.$

Wherein sgn () is a sign function, and threshold values λ is the number of discrete signal for

N.

4. parallel connection mixed type electric energy regulator expert's fault analysis according to claim 1 and diagnostic method, the concrete steps of said step (3) are:

(1) the fault sample signal after the denoising is carried out n layer WAVELET PACKET DECOMPOSITION, extract 2 of n layer ⁿIndividual each band signal characteristic;

(2) each coefficient of dissociation of wavelet packet is carried out reconstruct, in band limits separately, form the signal S after the reconstruct;

(3) ask the gross energy of each band signal;

(4) tectonic energy measure feature vector.

Images