CN103245893B - A kind of radial distribution layered distribution type ANN Fault Locating Method based on free-running frequency - Google Patents

Abstract

The invention provides a kind of radial distribution layered distribution type ANN Fault Locating Method utilizing free-running frequency, belong to Relay Protection Technology in Power System field.Utilize the nonlinear fitting ability of ANN to reflect this kind of mapping relations, realize localization of fault and the linear-elsatic buckling of radial distribution networks.When radial distribution networks generation singlephase earth fault, choose the residual voltage data of window during T/4 after fault, extract residual voltage free-running frequency and amplitude thereof by FFT, the input amendment using free-running frequency value as layered distribution type ANN model, first carries out localization of fault; Again using amplitude corresponding to free-running frequency as input amendment, carry out Fault branch identification, fault distance and place, trouble spot branch numbering export as it.The method overcome the shortcoming of wavefront fugitiveness in travelling wave ranging method, can effectively improve localization of fault antijamming capability, false voltage data fault-tolerant is strong.

Description

A kind of radial distribution layered distribution type ANN Fault Locating Method based on free-running frequency

Technical field

The present invention relates to Relay Protection Technology in Power System field, specifically a kind of radial distribution layered distribution type ANN Fault Locating Method based on free-running frequency.

Background technology

Consider the factors such as technology, economy and safety, China's medium voltage distribution network adopts small current neutral grounding mode mostly.Power distribution network is in electric system and user's contact the most directly link, it covers a wide range, singlephase earth fault probability is high, and technology and economic factors comprehensively determine multiple-limb radial distribution networks localization of fault its singularity, is done large quantifier elimination.Active fault position method and passive type fault position method can be divided into according to the difference of utilized signal.Active fault position method is by injecting characteristic frequency power signal to distribution network system, utilization movement or hard-wired signal detection apparatus come the position that detection failure occurs, as " S " injection method, add letter transfer function method etc., but its reliability is comparatively large by the impact of transition resistance, intermittent electric arc, and practical application effect is not very good.Passive type fault position method mainly utilizes break down front and back voltage, current signal feature of feeder line to realize localization of fault, wherein with impedance method and traveling wave method most representative.Impedance method is the judgement realizing abort situation based on the mapping relations between Fault loop impedance and fault distance, and because of distribution network voltage summation current transformer, to spread all over characteristic often inconsistent, cause larger range error can to the localization method based on impedance method.Meanwhile, for the labyrinth of multiple-limb power distribution network, impedance method such as effectively cannot to identify at the equivalent fault at electrical distance place.The time difference that traveling wave method utilizes transient state travelling wave to propagate between gauge point and trouble spot realizes localization of fault, but affect by multiple-limb power distribution network special construction, the accurate identification of second, trouble spot reflection wave head is more difficult, in addition more, the traveling wave method of the interference information source of power distribution network is to hardware and the more high factor of mutual inductor performance requirement, makes the effect of traveling wave method in multiple-limb power distribution network also not very good.

The present invention is radial distribution layered distribution type ANN (ArtificialNeuralNetwork, the artificial neural network) Fault Locating Method based on free-running frequency.When radial distribution networks generation singlephase earth fault, the capable ripple of false voltage is propagated along feeder line to both sides from trouble spot, and in wave impedance point of discontinuity generation catadioptric.Different branch breaks down, and the travel path that the capable ripple of its false voltage combines via different branch arrives bus bar side measuring end, and the free-running frequency of the fault transient voltage obtained by measuring end and amplitude distribution thereof are also not identical.There are mapping relations between the row propagation path of different branches combination and free-running frequency and amplitude distribution thereof.The nonlinear fitting ability that ANN can be utilized powerful, to reflect this kind of mapping relations, realizes localization of fault and the linear-elsatic buckling of radial distribution networks.After utilizing fault, during 1/4th power frequency periods, the residual voltage data of window add Chebyshev window and process, and extract residual voltage free-running frequency by FFT, the input amendment using free-running frequency value as layered distribution type ANN model, first carries out localization of fault; Again using amplitude corresponding to free-running frequency as input amendment attribute, carry out Fault branch identification, fault distance and place, trouble spot branch numbering are as its output sample attribute.Based on this, realize the localization of fault of radial distribution networks.

Summary of the invention

The object of the invention is to propose a kind of radial distribution layered distribution type ANN Fault Locating Method based on free-running frequency, overcome the shortcoming of wavefront fugitiveness in travelling wave ranging method, effectively improve localization of fault antijamming capability.

The radial distribution layered distribution type ANN Fault Locating Method that the present invention is based on free-running frequency is: when radial distribution networks generation singlephase earth fault, after utilizing fault, during 1/4th power frequency periods, the residual voltage data of window add Chebyshev window and process, residual voltage free-running frequency is extracted by FFT (FastFourierTransform, Fast Fourier Transform (FFT)); Input amendment using free-running frequency value as layered distribution type ANN (ArtificialNeuralNetwork, artificial neural network) model, carries out fault distance location; Using amplitude corresponding to free-running frequency as input amendment attribute, carry out Fault branch identification; Number as its output sample attribute using fault distance and place, trouble spot branch, realize the localization of fault of radial distribution networks.Concrete steps are as follows:

(1) when radial distribution networks generation singlephase earth fault, record three-phase voltage according to wave recording device and can obtain fault residual voltage, deducted corresponding temporal steady state voltage and obtain zero sequence transient fault voltage u ₀for:

$u_0\left(k\right)=\frac{1}{3}(u_A\left(k\right)+u_B\left(k\right)+u_C\left(k\right))---\left(2\right)$

In formula, u _a(k), u _b(k), u _ck () is respectively faulty line A, B, C three-phase voltage, k=1,2,3 ... N, N are sample sequence length;

(2) utilize T/4 short time-window bus bar side fault residual voltage data after multiple-limb radial distribution networks fault, carry out Chebyshev window extraction and carry out FFT conversion to residual voltage data, obtain the distribution of its free-running frequency, T is the cycle of power frequency amount.

(3) radial distribution networks localization of fault layered distribution type ANN: free-running frequency p in selecting step (2) is trained as follows ₁=(f _n1, f _n2, f _n3, f _n4, f _n5, f _n6, f _n7, f _n8) as the input amendment of ground floor neural network, f _n1, f _n2, f _n3, f _n4, f _n5, f _n6, f _n7, f _n8for Frequency point, E _n1, E _n2, E _n3, E _n4, E _n5, E _n6, E _n7, E _n8amplitude corresponding to Frequency point, its output vector y ₁=d _f, faults point distance; Free-running frequency p in selecting step (2) ₁the amplitude p of each correspondence ₂=(E _n1, E _n2, E _n3, E _n4, E _n5, E _n6, E _n7, E _n8) as the input amendment of second layer neural network, its output vector y ₂₁=1 or y ₂₁=2 and y ₂₂=3 or y ₂₂=4, faults point place branch; Number as its output sample attribute using fault distance and place, trouble spot branch;

(4) train the radial distribution networks localization of fault layered distribution type neural network obtained according to step (3), carry out localization of fault: output vector y ₁represent that trouble spot distance is d _f; When output vector is y ₂₁or y ₂₂when being respectively 1,2,3 or 4, represent that trouble spot is positioned at respective branches L ₁₂, L ₁₅, L ₁₃or L ₁₄on.

In the present invention, during to the ANN model training of radial distribution networks localization of fault layered distribution type, choosing of input amendment is carried out in conjunction with following fault condition: along outgoing feeder ₁choose trouble spot, fault distance change step is 200m; Fault resistance R gets 20 Ω, 100 Ω, 500 Ω respectively; The initial phase angle of fault gets 30 °, 60 °, 90 ° respectively.

In the present invention, measure power distribution network bus bar side residual voltage, the length of data window is the fault just surely rear 5ms of instantaneous blaze, and sample frequency is 100kHz.

Principle of the present invention is:

1, free-running frequency signature analysis

When radial distribution networks generation singlephase earth fault, the capable ripple of false voltage is propagated along feeder line to both sides from trouble spot, and in wave impedance point of discontinuity generation catadioptric.Fault occurs in different branch, the capable ripple of false voltage arrives the measuring end of transformer substation side via the travel path that different branch combines, the free-running frequency distribution of the fault transient voltage obtained by measuring end is also not identical, there are mapping relations between the row propagation path of different branches combination and free-running frequency distribute.There are mapping relations in free-running frequency distribution with abort situation.

Fig. 2 gives the fault component zero-sequence network of direct distribution lines distribution parameter equivalence, wherein, and R ₀, L ₀, G ₀and C ₀be respectively the resistance of circuit unit length positive sequence, inductance, conductance and electric capacity; Z _sfor source impedance, Z _mfor load side impedance; F is trouble spot; Z _ffor fault point impedance; U _ffor F point place voltage before fault; V is fault traveling wave velocity of wave; D is direct distribution lines total length, d _ffor trouble spot is to the distance of mains side.

The initial row ripple road direction both sides along the line that fault produces are propagated, and the superposition of the row ripple of what measuring end was experienced is measuring end M and trouble spot, end multiple reflections, shows in frequency to be a series of high fdrequency component with fixed frequency.

Measuring end M holds the faults free-running frequency obtained to meet:

$1-{\mathrm{\β}}_M{\mathrm{\β}}_F{e}^{-2\mathrm{\ω}{\mathrm{\τ}}_1}=0---\left(2\right)$

The free-running frequency of reflection total track length meets:

$1-{\mathrm{\β}}_M{\mathrm{\β}}_N{e}^{-2\mathrm{\ω}{\mathrm{\τ}}_2}=0---\left(3\right)$

In formula (2) and (3), β _mfor measuring end M reflection coefficient, β _ffor trouble spot reflection coefficient, ω is angular frequency, the time propagating into measuring end M from trouble spot for row ripple, (wherein, v was fault traveling wave velocity of propagation, d _ffor trouble spot is to the distance of mains side).

Known, there is infinite multiple solution formula (2) and (3), utilize Euler's formula to be translated into exponential form, can be expressed as

e ^2ωτ＝β _Mβ _Fe ^2kπik＝0,±1,±2,...(4)

And then the expression that can obtain row ripple free-running frequency ω is

${\mathrm{\ω}}_k=\frac{1}{\mathrm{\τ}}\ln \left(\sqrt{\left|{\mathrm{\β}}_M{\mathrm{\β}}_F\right|}\right)+i\frac{{\mathrm{\θ}}_M+{\mathrm{\θ}}_F+2\mathrm{k\π}}{2\mathrm{\τ}}---\left(5\right)$

In formula, θ _m=arg β _m, θ _f=arg β _f; Wherein θ _mfor the reflection angle of measuring end, θ _ffor trouble spot reflection angle.

In formula (5), real part represents the attenuation degree of free-running frequency, and its imaginary part is the angular frequency of this free-running frequency, then the free-running frequency of fault traveling wave can be expressed as:

$f_k=\frac{\mathrm{Im\λ}}{2\mathrm{\π}}=\frac{{\mathrm{\θ}}_M+{\mathrm{\θ}}_F+2\mathrm{k\π}}{4\mathrm{\π\τ}},k=0,\±1,\±2,...\left(6\right)$

Under extreme boundary situation, when system side is open circuit, when trouble spot is short circuit, β _mand β _fbe real number, and β _mβ _fwhen=-1, now, θ _m=π, θ _f=0, the major component of row ripple free-running frequency is:

$f=\frac{v}{4d_f}---\left(7\right)$

In like manner, when circuit two ends are open circuit, i.e. β _mβ _nwhen=1, wherein β _nfor feeder terminal reflection coefficient, the major component of row ripple free-running frequency is:

$f=\frac{v}{2d}---\left(8\right)$

In formula (7), (8), v is fault traveling wave velocity of propagation, d _ffor trouble spot is to the distance of mains side, d is direct distribution lines total length.

In actual conditions, consider that bus-bar system and the step-down of distribution end become the existence of stray capacitance and trouble spot transition resistance over the ground, when utilizing formula (7) or (8) to carry out suspected fault voltage traveling wave free-running frequency, there is certain deviation.

As the above analysis, the failed row wave frequency that measuring end obtains is all relevant with trouble spot reflection angle with the reflection angle of measuring end, and there are mapping relations between fault traveling wave free-running frequency major component and trouble spot distance, free-running frequency (distribution) well reflects abort situation.

2, the radial distribution layered distribution type ANN of free-running frequency is utilized to carry out localization of fault

For the radial distribution of multiple-limb, fault occurs in different branch, the capable ripple of false voltage arrives the measuring end of transformer substation side via the travel path that different branch combines, the free-running frequency distribution of the fault transient voltage obtained by measuring end is different with amplitude, there are mapping relations between the row propagation path of different branches combination and free-running frequency distribute.Wherein, the range information reaction between trouble spot to measuring end is in the free-running frequency distribution of the capable ripple of false voltage, and fault branch information response is in free-running frequency and amplitude thereof.The nonlinear fitting ability utilizing ANN powerful, can reflect this kind of mapping relations.In order to improve the learning efficiency of neural network, optimizing the structure of BP model, adopting layered distribution type neural network model.

When radial distribution networks system generation singlephase earth fault, arbitrary fault component all contains the information of abort situation, but different fault components is different from the degree of strength of the mapping relations between abort situation, therefore select to distribute as the characteristic quantity of neural network with the free-running frequency of the strong mapping relations of abort situation, input amendment can be reduced, convergence speedup, and then reach accurate fault location.As shown in Figure 4, wherein ground floor ANN is used for localization of fault, and its input amendment attribute is free-running frequency distribution, and second layer ANN is used for Fault branch identification, and its input amendment attribute is amplitude corresponding to free-running frequency.Utilize ground floor localization of fault ANN to carry out localization of fault, when exporting certain section on fault distance trunk and without branch, then Output rusults is position of failure point; When exporting fault distance and being positioned at certain section on trunk and having branch, then need to utilize second layer linear-elsatic buckling ANN to carry out Fault branch identification according to Output rusults.

The present invention compared with prior art tool has the following advantages:

(1) this method utilizes the capable ripple of false voltage to propagate to both sides from trouble spot along outgoing, and in wave impedance point of discontinuity generation catadioptric, failed row wave trajectory and free-running frequency distribution also exist mapping relations; Free-running frequency range finding be utilize measuring end to experience fault traveling wave wave head repeatedly catadioptric formed high fdrequency component, it does not rely on the detection of one or two wavefront, thus overcomes the shortcoming of wavefront fugitiveness in travelling wave ranging method.

(2) this method adopts the nonlinear fitting ability that ANN is powerful, can effectively improve localization of fault antijamming capability.

(3) this method does not affect by transition resistance, the initial phase angle of fault to a certain extent, and false voltage data fault-tolerant is strong.

Accompanying drawing explanation

Fig. 1 is multiple-limb radial distribution networks system architecture schematic diagram of the present invention; In figure, T is main-transformer, T _zfor Z-shaped transformer, L is arc suppression coil, and R is the damping resistance of arc suppression coil, and K is disconnector, and M point is the measuring end of the fault traveling wave being positioned at bus place, C _efor bus-bar system equivalent capacity over the ground, f is trouble spot.

Fig. 2 is the fault component zero-sequence network figure of direct distribution lines distribution parameter equivalence of the present invention, in figure, and R ₀, L ₀, G ₀and C ₀for the resistance of circuit unit length positive sequence, inductance, conductance and electric capacity; Z _sfor source impedance, Z _mfor load side impedance; F is trouble spot; Z _ffor fault point impedance; U _ffor F point place voltage before fault; V is fault traveling wave velocity of wave; D is direct distribution lines total length, d _ffor trouble spot is to the distance of mains side;

Fig. 3 is after power distribution network generation singlephase earth fault of the present invention, the residual voltage time domain beamformer that measuring end M detects; In figure, horizontal axis representing time, unit is ms, and the longitudinal axis represents residual voltage size, and unit is kV;

Fig. 4 is after power distribution network generation singlephase earth fault of the present invention, intercepts 4.096ms data and carry out adding Chebyshev window FFT conversion, the free-running frequency distribution plan obtained from window data during 5ms; Transverse axis represents frequency, and unit is kHz, and the longitudinal axis represents frequency amplitude size, and unit is kV.

Fig. 5 is radiation Distribution Network Failure of the present invention location layered distribution type neural network model figure.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is further elaborated.

When radial distribution networks generation singlephase earth fault, record three-phase voltage according to wave recording device, pass through formula obtain zero sequence transient fault voltage u ₀; After utilizing fault, during 1/4th power frequency periods, the residual voltage data of window add Chebyshev window and process, and extract residual voltage free-running frequency by FFT; Build radiation Distribution Network Failure location layered distribution type neural network (ANN) model, with free-running frequency value p ₁=(f _n1, f _n2, f _n3, f _n4, f _n5, f _n6, f _n7, f _n8) as the input amendment of the ground floor neural network of layered distribution type ANN model, obtain output vector y ₁, according to y ₁=d _fcarry out fault distance location; With free-running frequency p ₁corresponding amplitude p ₂=(E _n1, E _n2, E _n3, E _n4, E _n5, E _n6, E _n7, E _n8) as the input amendment of the second layer neural network of layered distribution type ANN model, obtain output vector y ₂₁=1 or y ₂₁=2 and y ₂₂=3 or y ₂₂=4, carry out Fault branch identification; Number as its output sample attribute using fault distance and place, trouble spot branch.The localization method of this patent is first according to y ₁=d _fdetermine fault distance, as output vector y ₂₁or y ₂₂when being respectively 1,2,3 or 4, represent that trouble spot is positioned at respective branches L ₁₂, L ₁₅, L ₁₃or L ₁₄on.

Embodiment 1: for multiple-limb electricity distribution network model as shown in Figure 1, this system neutral, through grounding through arc, arranges fault outgoing feeder on the right side of bus ₁, arrange on the left of bus and perfect outgoing feeder ₂and feeder ₃.Feed out circuit feeder ₁comprise 5 branches, be respectively L ₁₁, L ₁₂, L ₁₃, L ₁₄, and L ₁₅; Feed out circuit feeder ₂circuit is fed out for directly joining; Feed out circuit feeder ₃comprise 3 branches, be respectively L ₃₁, L ₃₂and L ₃₃.T in this power distribution network is main-transformer, and voltage ratio is 110kV/35kV, and connection set is YN/Y; T _zit is Z-shaped transformer; L is arc suppression coil; R is the damping resistance of arc suppression coil.Arc suppression coil is by disconnector K switching.M point is the measuring end of the fault traveling wave being positioned at bus place, and bus-bar system is equivalent capacity C over the ground _ebe 0.0001 μ F, the equivalent capacity over the ground that load side step-down becomes is as shown in frame empty in Fig. 1.Feed out circuit feeder ₁branched line L ₁₃singlephase earth fault occurs, and trouble spot f distance bus bar side 5.9km, transition resistance is 0.01 Ω, fault angle θ=90 °, and fault moment is 0.084s, window T during data _s=5ms, sample frequency is 100kHz.

(1), after power distribution network generation singlephase earth fault, measuring end M detects residual voltage time domain waveform as shown in Figure 3.When choosing 5ms after startup, window data carry out adding Chebyshev window FFT conversion, obtain free-running frequency distribution as shown in Figure 4.

(2) corresponding free-running frequency p is extracted ₁=(11.32,10.8,16.01,18.94,22.25,26.15,29.5,38.2), ground floor localization of fault is carried out in input as shown in Figure 5 radial distribution layered distribution type ANN model, exports y ₁=5.6188; Recycling free-running frequency p ₁corresponding amplitude p ₂=(15.7,5.28,0.51,1.32,0.27,1.17,0.384,0.005), in input radiation shape distribution layered distribution type ANN second layer localization of fault neural network, Output rusults y ₂₂=3.0010.

(3) according to acquired results y in (2) ₁=5.6188, determine that fault is positioned at distance bus bar side 5.6188km place, physical fault distance is 5.9km, its error ε _r=4.77%; By y ₂₂=3.0010, determine that fault is positioned at and feed out circuit L ₁₃on, its result is consistent with actual conditions, and locating effect is good.

Embodiment 2: the same model adopted in embodiment 1, feeds out circuit feeder ₁branched line L ₁₅singlephase earth fault occurs, and trouble spot f distance bus bar side 2.5km, transition resistance is 0.01 Ω, fault angle θ=90 °, and fault moment is 0.084s, window T during data _s=5ms, sample frequency is 100kHz.

Extract corresponding free-running frequency p ₁=(31.6,11.9,16.2,17,20.1,25.78,41.2,46.9), in input radiation shape distribution layered distribution type ANN ground floor localization of fault neural network, export y ₁=2.4053; Recycling free-running frequency p ₁corresponding amplitude p ₂=(0.23,8.12,0.209,0.188,0.26,0.08,0.065,0.02), in input radiation shape distribution layered distribution type ANN second layer localization of fault neural network, Output rusults y ₂₁=1.9981.

According to the Output rusults y of neural network ₁=2.4053, determine that fault is positioned at distance bus bar side 2.4053km place, physical fault distance is 2.5km, its error ε _r=3.79%; By y ₂₁=1.9981, determine that fault is positioned at and feed out circuit L ₁₅on, its result is consistent with actual conditions, and locating effect is good.

By reference to the accompanying drawings embodiments of the present invention are illustrated above, but the present invention is not limited to above-mentioned embodiment, in the ken that those skilled in the art possess, can also makes a variety of changes under the prerequisite not departing from present inventive concept.

Claims (3)

1. the radial distribution networks layered distribution type ANN Fault Locating Method based on free-running frequency, it is characterized in that: when radial distribution networks generation singlephase earth fault, after utilizing fault, during 1/4th power frequency periods, the residual voltage data of window add Chebyshev window and process, and extract residual voltage free-running frequency by FFT; Input amendment using free-running frequency value as layered distribution type ANN model, carries out fault distance location; Using amplitude corresponding to free-running frequency as input amendment attribute, carry out Fault branch identification; Number as its output sample attribute using fault distance and place, trouble spot branch, realize the localization of fault of radial distribution networks;

Concrete steps are as follows:

(1) when radial distribution networks generation singlephase earth fault, record three-phase voltage according to wave recording device and can obtain fault residual voltage, deducted corresponding temporal steady state voltage and obtain zero sequence transient fault voltage u ₀for:

$u_0\left(k\right)=\frac{1}{3}(u_A\left(k\right)+u_B\left(k\right)+u_C\left(k\right))---\left(1\right)$

In formula, u _a(k), u _b(k), u _ck () is respectively faulty line A, B, C three-phase voltage, k=1,2,3 ... N, N are sample sequence length;

(2) utilize T/4 short time-window bus bar side fault residual voltage data after multiple-limb radial distribution networks fault, carry out Chebyshev window extraction and carry out FFT conversion to residual voltage data, obtain the distribution of its free-running frequency, T is the cycle of power frequency amount;

(3) build radial distribution networks localization of fault layered distribution type ANN, and train as follows: free-running frequency p in selecting step (2) ₁=(f _n1, f _n2, f _n3, f _n4, f _n5, f _n6, f _n7, f _n8) as the input amendment of ground floor ANN, its output vector y ₁=d _f, faults point distance; Free-running frequency p in selecting step (2) ₁the amplitude p of each correspondence ₂=(E _n1, E _n2, E _n3, E _n4, E _n5, E _n6, E _n7, E _n8) as the input amendment of second layer ANN, f _n1, f _n2, f _n3, f _n4, f _n5, f _n6, f _n7, f _n8for Frequency point, E _n1, E _n2, E _n3, E _n4, E _n5, E _n6, E _n7, E _n8amplitude corresponding to Frequency point, its output vector y ₂₁=1 or y ₂₁=2 and y ₂₂=3 or y ₂₂=4, faults point place branch; Number as its output sample attribute using fault distance and place, trouble spot branch;

(4) train the radial distribution networks localization of fault layered distribution type ANN obtained according to step (3), carry out localization of fault: output vector y ₁expression fault distance is d _f; As output vector y ₂₁or y ₂₂when being respectively 1,2,3 or 4, represent that trouble spot is positioned at respective branches L ₁₂, L ₁₅, L ₁₃or L ₁₄on.

2. the radial distribution networks layered distribution type ANN Fault Locating Method based on free-running frequency according to claim 1, it is characterized in that: during to the ANN model training of radial distribution networks localization of fault layered distribution type, choosing of input amendment is carried out in conjunction with following fault condition: along outgoing feeder ₁choose trouble spot, fault distance change step gets 200m; Fault resistance R gets 20 Ω, 100 Ω, 500 Ω respectively; The initial phase angle of fault gets 30 °, 60 °, 90 ° respectively.

3. the radial distribution networks layered distribution type ANN Fault Locating Method based on free-running frequency according to claim 1, is characterized in that: measure power distribution network bus bar side voltage time, time window length be 5ms after fault, sample frequency is 100kHz.