CN1896756A - Cable-fault on-line distance measurement based on artifical nerve network mode - Google Patents

Abstract

The invention relates to an online distance measurement method of power cable malfunction bases on manual neural network model, to ensure the location of the power cable malfunction. The method's connects the three-phase current analog signals which is from the current transformer and the voltage transformer in both ends of the cable, zero order current analog signals and three-phase voltage analog signals, zero order voltage analog signals with the low-pass filter circuit, transmitting the handled current and voltage analog signals to the frequency lock phase lock sampling module, the current and voltage digital signal which is exported from the frequency lock phase lock are used to the malfunction startup module to start the malfunction, it is used for cable ANN model training module to train the cable ANN model , it is used for cable malfunction distance measurement calculating module to calculate the distance of the power cable malfunction, ascertaining the location of the power cable malfunction accurately.

Description

Based on the online distance-finding method of the cable fault of artificial nerve network model

Technical field

The present invention relates to cable fault localization method, the online distance-finding method of particularly a kind of cable fault based on artificial nerve network model is used for determining of power cable fault position.

Technical background

Existing power system failure location (also claiming fault localization) method mainly can be divided into two big classes both at home and abroad: traveling wave method and impedance method.This two big class all comprises single-ended method and both-end method again according to the difference of Data Source angle.

The principle of TRAVELING WAVE FAULT LOCATION is to arrive the bus back reflection to the trouble spot according to the row ripple, is arrived the mistiming of bus again by the trouble spot reflection, or arrives a kind of algorithm that mistiming of two side bus finds range according to going the initial wave head of ripple.

Wherein the single-ended traveling wave method is based on a kind of distance-finding method of single-ended quantity of information, and the key of single-ended traveling wave range finding is to obtain accurately that capable ripple arrives measuring junction for the first time and it reflects back into the mistiming of measuring junction from the trouble spot, and comprises the extraction of fault traveling wave component.Capable ripple one-end fault ranging algorithm commonly used has differentiation, correlation method, matched filter method and predominant frequency method.Because the row ripple is in the catadioptric situation more complicated (after as row ripple arrival trouble spot reflection can take place also can be refracted to side bus is got on by the trouble spot) of characteristic impedance variation place, non-fault line is not " endless ", reflect traveling-wave component in the past behind certain hour by measurement point, again can be from measurement point bounce back faulty line etc., make travelling wave analysis and utilize the accurate fault localization of single-ended traveling wave that big difficulty is arranged.So the more of research now is the both-end traveling wave method.The key of both-end traveling wave method is the time that the capable ripple of curtage arrives the circuit two ends under the accurately record, and error is in several μ s, to guarantee that the fault localization error is in hundreds of rice.But it needs special-purpose unit lock in time.

In actual transmission line of electricity, because inhomogeneous, the line parameter circuit value of line construction variations, interchanging method difference, power transmission line ground resistivity along the line is with problems such as the variation of frequency, capable wave dispersions, the characteristics that make fault produce the row ripple can not be fully utilized, travelling wave analysis and research is difficulty relatively, and very high to the requirement of device.

The impedance fault telemetry is to come a kind of method of suspected fault impedance point distance by finding the solution line impedance value from the trouble spot to the measurement point.When setting up transmission line malfunction impedance range finding model, usually fault distance is taken into account as a circuit parameter, by the solving circuit equation, obtain the impedance distance.

The impedance ratio juris is the hypothesis of uniform line based on transmission line of electricity, i.e. assumed fault impedance loop or reactance is directly proportional with the distance of measurement point to the trouble spot.During fault, measurement mechanism is started by starting element, parameters such as the voltage and current when recording fault, and then calculate the impedance of fault loop.Because line length is directly proportional with impedance, therefore can obtain by the distance of device installation place to the trouble spot.

The calculating of fault impedance can use the information at circuit two ends to carry out, and also can only utilize an end metrical information to carry out approximate processing.Single-ended measurement is only used voltage, the current measurement value of circuit one side apart from algorithm, can't overcome the influence of transition resistance in theory, and need do certain hypothesis in location algorithm, so its measuring accuracy is difficult to guarantee under many circumstances.Both-end measures can't eliminate the influence of following factors to distance accuracy fully apart from algorithm: circuit model, line parameter circuit value imbalance, line parameter circuit value are inaccurate, load current, synchro measure precision and fundametal compoment are extracted precision.

Summary of the invention

The purpose of this invention is to provide the online distance-finding method of a kind of cable fault, carry out the fault localization of power cable, to determine the position of failure point of power cable more exactly based on artificial nerve network model.

The online distance-finding method of cable fault based on artificial nerve network model of the present invention is:

Will be from cable two ends current transformer and the next three-phase current simulating signal i of voltage transformer (VT) _An(i _Bn, i _Cn), zero-sequence current simulating signal i _0nWith three-phase voltage simulating signal v _An(v _Bn, v _Cn), residual voltage simulating signal v _0nInsert low-pass filter circuit, electric current, voltage analog signal after low-pass filter circuit is handled are sent into the phase-locked sampling module of frequency locking, the electric current of exporting, voltage digital signal i ' from the phase-locked sampling module of frequency locking _An(i ' _Bn, i ' _Cn), i ' _0n, v ' _An(v ' _Bn, v ' _Cn), v ' _0nBe used for fault initiating module, cable ANN (artificial neural network) model training module and cable fault localization computing module.

The fault initiating module is made up of short trouble pretrigger module, single-phase earthing pretrigger module, short trouble affirmation module, single-phase earthing affirmation module and ground path judge module, with the current digital signal i ' that exports in the phase-locked sampling module of frequency locking _An(i ' _Bn, i ' _Cn) insert the adaptive sine wave filter of short trouble pretrigger module, carry out the short trouble pretrigger; The residual voltage digital signal v ' that the phase-locked sampling module of frequency locking is come out _0nInsert the adaptive sine wave filter of single-phase earthing pretrigger module, carry out the singlephase earth fault pretrigger.

If short trouble pretrigger module output short-circuit fault pre-actuation signal is with voltage, the current digital signal v ' of the phase-locked sampling module output of frequency locking _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) insert the sinusoidal approximation process device that short trouble is confirmed module, carry out short trouble and confirm, confirm back output short-circuit fault recognition signal.

If single-phase earthing pretrigger module output singlephase earth fault pre-actuation signal is with the three-phase voltage and the residual voltage digital signal v ' of the phase-locked sampling module output of frequency locking _An(v ' _Bn, v ' _Cn), v ' _0nInsert single-phase earthing and confirm the sinusoidal approximation process device of module, carry out singlephase earth fault and confirm, confirm back output singlephase earth fault confirmation signal, the residual voltage and the zero-sequence current digital signal v ' of the phase-locked sampling module output of frequency locking _0n, i ' _0nInsert the sinusoidal approximation process device of ground path judge module, carry out ground path and judge, the output ground path is judged signal.

After fault initiating module output short-circuit fault recognition signal or ground path are judged signal, with voltage, the current digital signal v ' at the cable two ends of the phase-locked sampling module output of frequency locking _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) insert six cable ANN (artificial neural network) model, carry out determining of Method of Cable Trouble Point position.

If the fault initiating module is when output short-circuit fault pre-actuation signal and/or singlephase earth fault pre-actuation signal, with voltage, the current digital signal v ' at the cable two ends of the phase-locked sampling module output of frequency locking _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) insert six cable ANN (artificial neural network) model, carry out cable ANN model training.

Described adaptive sine wave filter is by one and current digital signal i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches current digital signal i ' _An(i ' _Bn, i ' _Cn) constitute, its mathematical expression is:

y(t)＝Acos(ωt)+Bsin(ωt)

e(t)＝y(t)-i(t)

In order to regulate corrected parameter A and B, carry out following correction algorithm:

A′＝A-μe(t)cos(ωt)

B′＝B-μe(t)sin(ωt)

A ', B ' are the modified values of adaptive sine filter parameter in the formula, and μ (μ＞0) is the algorithm convergence factor.

Short trouble pretrigger method is: current digital signal i ' _An(i ' _Bn, i ' _Cn) and the wave filter sinusoidal signal between error e by adaptive sine wave filter output, and with error e and error definite value e _Set1Compare, if e＞e _Set1, and the sum of errors ∑ | e (j) | greater than error definite value e _Set2, output short-circuit fault pre-actuation signal then, on the contrary single-phase earthing pretrigger module then entered.

Singlephase earth fault pretrigger method is: residual voltage digital signal v ' _0nAnd the error e between the wave filter sinusoidal signal ₀By the output of adaptive sine wave filter, and with error e ₀With error definite value e _Set3Compare, if e ₀＞e _Set3, and the sum of errors ∑ | e ₀(j) | greater than error definite value e _Set4, then export the single-phase earthing pre-actuation signal, otherwise then enter cable ANN (artificial neural network) model training.

Sinusoidal approximation process device by one with voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches voltage, current digital signal v ' in a time domain interval _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) constitute, in sinusoidal approximation process device, u (t) is voltage or current digital input signal,

y(t)＝A(t)cosωt+B(t)sinωt

A(t)＝A ₀+A ₁t+A ₂t ²+…+A _Nt ^N

B(t)＝B ₀+B ₁t+B ₂t ²+…+B _Nt ^N

A ₀, A ₁..., A _NAnd B ₀, B ₁..., B _NIt is the parameter of function A (t), B (t).

Note $J=\underset{i=1}{\overset{I}{\mathrm{\Σ}}}e{\left(t_i\right)}^2=\underset{i=1}{\overset{I}{\mathrm{\Σ}}}{(u\left(t_i\right)-y\left(t_i\right))}^2,W^k={\left[\begin{array}{cccccc}{A}_0^k& \·\·\·& A_N^k& B_0^k& \·\·\·& B_N^k\end{array}\right]}^T,$ N is the exponent number of polynomial function, and I is that the calculating in the section is counted, and k is for approaching calculation times.

The correction algorithm of sinusoidal approximation process device is

$W^{k+1}=W^k-\mathrm{\μ}\frac{\∂J}{\∂W^k}$ μ (μ＞0) is the algorithm convergence factor.

At J＜J _MinThe time, function A (t), B (t) after sinusoidal approximation process device output approaches.

The short trouble confirmation method is: calculate $v_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ With $i_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) instantaneous amplitude v _m, i _m, if electric current instantaneous amplitude i _mIncrement i _m(t+ Δ t)-i _m(t) greater than current ration i _Set1, instantaneous amplitude i _m(t+ Δ t) is greater than current ration i _Set2, and voltage instantaneous amplitude v _mIncrement v _m(t)-v _m(t+ Δ t) is greater than voltage definite value v _Set1, instantaneous amplitude v _m(t+ Δ t) is less than voltage definite value v _Set2, output short-circuit fault recognition signal then.

The singlephase earth fault confirmation method is: calculate $v_{\mathrm{mA}}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ (v _MB(t), v _MC(t)) and $v_{m0}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain three-phase voltage and residual voltage digital signal v ' _An(v ' _Bn, v ' _Cn), v ' _0nInstantaneous amplitude v _m, v _M0, if residual voltage instantaneous amplitude v _M0Increment v _M0(t+ Δ t)-v _M0(t) greater than residual voltage definite value v _Set4, instantaneous amplitude v _M0(t+ Δ t) is greater than residual voltage definite value v _Set5, and the phase voltage instantaneous amplitude v in the three-phase voltage _m(t+ Δ t) is less than voltage definite value v _Set3, then export the single-phase earthing confirmation signal.

The ground path determination methods is: calculate  _v(t)=artgA (t)/B (t)) and  _i(t)=artg (A (t)/B (t)) can obtain residual voltage and zero-sequence current digital signal v ' _0n, i ' _0nInstantaneous phase  _v(t),  _i(t), if the instantaneous phase  of zero-sequence current _i(t) be ahead of the instantaneous phase  of residual voltage _v(t), then export ground path and judge signal.

The Method of Cable Trouble Point method for determining position is: for short trouble, change the cable length parameter l, make the output v of cable ANN model of the fault phase at cable two ends _l ¹With v _l ²The difference minimum, this moment the cable length parameter l be exactly the distance of check point to the trouble spot; For single-phase earthing, change the cable length parameter l, make the output sum of threephase cable ANN (artificial neural network) model at cable two ends, i.e. residual voltage v _0l ¹With v _0l ²The difference minimum, this moment the cable length parameter l be exactly the distance of check point to the trouble spot.

Cable ANN (artificial neural network) model training method is: each cable ANN model inserts a phase voltage and three-phase current, the single order that calculates input voltage and electric current is to the n order derivative, and with these derivatives also as the input of cable ANN model, the output of cable ANN model is voltage v _l, cable ANN model comprises the cable length parameter l, the output v of the cable ANN model at cable two ends _Al ¹(v _Bl ¹, v _Cl ¹) and v _Al ²(v _Bl ², v _Cl ²) difference be the training error e of cable ANN model, to the input of different voltage and current, change the cable length parameter, regulate the weight coefficient W of cable ANN model _V1, W _IA1, W _IB1, W _IC1... W _Vn, W _IAn, W _IBn, W _ICn, make the training error e of cable ANN model ²Less than e _Set5

The online distance-finding method of cable fault of the present invention, by making up artificial neural network (ANN) model of cable fault localization physical object, physical object with this replacement cables fault localization, the ANN model of cable fault localization physical object can be under the cable normal operation, and current/voltage synchronized sampling value and its all-order derivative by the cable two ends come learning training.Owing to adopt the real-time sampling data that the ANN model is constantly trained, so the ANN model has comprised fault moment cable data accurately, the influence that not changed by system and cable data.When cable fault, ANN model according to the cable fault localization physical object of training in real time, utilize the electric current of cable one end, the voltage that the voltage sample value is calculated the cable other end, regulate the cable length parameter, make the voltage error minimum of its calculating, obtain the distance of check point with this,, can determine the position of failure point of power cable more exactly so utilize the inventive method that the fault of power cable is found range to the trouble spot.

Description of drawings

Now in conjunction with the accompanying drawings the present invention is described in further detail.

Fig. 1 is the online distance-finding method block diagram of the cable fault based on artificial nerve network model of the present invention;

Fig. 2 is fault initiating method synoptic diagram among the present invention;

Fig. 3 is short trouble pretrigger method synoptic diagram among the present invention;

Fig. 4 is singlephase earth fault pretrigger method synoptic diagram among the present invention;

Fig. 5 is an adaptive sine filter construction synoptic diagram;

Fig. 6 is short circuit fault recognition method synoptic diagram among the present invention;

Fig. 7 is singlephase earth fault confirmation method synoptic diagram among the present invention;

Fig. 8 is ground path determination methods synoptic diagram among the present invention;

Fig. 9 is sinusoidal approximation process method synoptic diagram;

Figure 10 is cable ANN model training method synoptic diagram among the present invention;

Figure 11 is cable ANN model structure synoptic diagram among the present invention;

Figure 12 is cable fault localization computing method synoptic diagram among the present invention;

Figure 13 is the structured flowchart of the online distance-finding method of cable fault based on artificial nerve network model of the present invention.

Embodiment

As shown in figure 1 to figure 13, should be based on the online distance-finding method of the cable fault of artificial nerve network model:

1, will be from cable two ends current transformer and the next three-phase current simulating signal i of voltage transformer (VT) _An(i _Bn, i _Cn), zero-sequence current simulating signal i _0nWith three-phase voltage simulating signal v _An(v _Bn, v _Cn), residual voltage simulating signal v _0nInsert low-pass filter circuit 11, electric current, voltage analog signal after low-pass filter circuit 11 is handled are sent into the phase-locked sampling module 12 of frequency locking, electric current, the voltage digital signal i ' of output from the phase-locked sampling module 12 of frequency locking _An(i ' _Bn, i ' _Cn), i ' _0n, v ' _An(v ' _Bn, v ' _Cn), v ' _0nBe used for fault initiating module 2, cable ANN (artificial neural network) model training module 3 and cable fault localization computing module 4.

2, fault initiating module 2 is made up of short trouble pretrigger module 21, single-phase earthing pretrigger module 22, short trouble affirmation module 23, single-phase earthing affirmation module 24 and ground path judge module 25.

1), the current digital signal i ' that will come out from the phase-locked sampling module 12 of frequency locking _An(i ' _Bn, i ' _Cn) insert the adaptive sine wave filter 51 of short trouble pretrigger module 21, adaptive sine wave filter 51 by one with current digital signal i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches current digital signal i ' _An(i ' _Bn, i ' _Cn) constitute, its mathematical expression is:

y(t)＝Acos(ωt)+Bsin(ωt)

e(t)＝y(t)-i(t)

In order to regulate corrected parameter A and B, carry out following correction algorithm

A′＝A-μe(t)cos(ωt)

B′＝B-μe(t)sin(ωt)

A ', B ' are the modified values of adaptive sine wave filter 51 parameters in the formula, and μ (μ＞0) is the algorithm convergence factor.

Current digital signal i ' _An(i ' _Bn, i ' _Cn) and the wave filter sinusoidal signal between error e by adaptive sine wave filter 51 output, and with error e and error definite value e _Set1Compare, if e＞e _Set1, and the sum of errors ∑ | e (j) | greater than error definite value e _Set2, output short-circuit fault pre-actuation signal then, on the contrary single-phase earthing pretrigger module 22 then entered.

2), the residual voltage digital signal v ' that will come out from the phase-locked sampling module 12 of frequency locking _0nInsert the adaptive sine wave filter 51 of single-phase earthing pretrigger module 22, residual voltage digital signal v ' _0nAnd the error e between the wave filter sinusoidal signal ₀By 51 outputs of adaptive sine wave filter, and with error e ₀With error definite value e _Set3Compare, if e ₀＞e _Set3, and the sum of errors ∑ | e ₀(j) | greater than error definite value e _Set4, then export the singlephase earth fault pre-actuation signal, otherwise then enter cable ANN model training 3.

3), after the output short-circuit fault pre-actuation signal, will be from voltage, the current digital signal v ' of phase-locked sampling module 12 outputs of frequency locking _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) insert the sinusoidal approximation process device 52 that short trouble is confirmed module 23, sinusoidal approximation process device 52 by one with voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches voltage, current digital signal v ' in a time domain interval _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) constitute, in sinusoidal approximation process device 52, u (t) is voltage or current digital input signal,

y(t)＝A(t)cosωt+B(t)sinωt

A(t)＝A ₀+A ₁t+A ₂t ²+…+A _Nt ^N

B(t)＝B ₀+B ₁t+B ₂t ²+…+B _Nt ^N

A ₀, A ₁..., A _NAnd B ₀, B ₁..., B _NIt is the parameter of function A (t), B (t).

Note $J=\underset{i=1}{\overset{I}{\mathrm{\Σ}}}e{\left(t_i\right)}^2=\underset{i=1}{\overset{I}{\mathrm{\Σ}}}{(u\left(t_i\right)-y\left(t_i\right))}^2,W^k={\left[\begin{array}{cccccc}{A}_0^k& \·\·\·& A_N^k& B_0^k& \·\·\·& B_N^k\end{array}\right]}^T,$ N is the exponent number of polynomial function, and I is that the calculating in the section is counted, and k is for approaching calculation times.

The correction algorithm of sinusoidal approximation process device 52 is

$W^{k+1}=W^k-\mathrm{\μ}\frac{\∂J}{\∂W^k}$ μ (μ＞0) is the algorithm convergence factor.

At J＜J _MmThe time, function A (t), B (t) after sinusoidal approximation process device 52 outputs approach calculate $v_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ With $i_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) instantaneous amplitude v _m, i _m, if electric current instantaneous amplitude i _mIncrement i _m(t+ Δ t)-i _m(t) greater than current ration i _Set1, instantaneous amplitude i _m(t+ Δ t) is greater than current ration i _Set2, and voltage instantaneous amplitude v _mIncrement v _m(t)-v _m(t+ Δ t) is greater than voltage definite value v _Set1, instantaneous amplitude v _m(t+ Δ t) is less than voltage definite value v _Set2, then output short-circuit fault recognition signal carries out the range finding of short trouble and calculates.

4), after the output singlephase earth fault pre-actuation signal, will be from the three-phase voltage and the residual voltage digital signal v ' of phase-locked sampling module 12 outputs of frequency locking _An(v ' _Bn, v ' _Cn), v ' _0nInsert single-phase earthing and confirm the sinusoidal approximation process device 52 of module 24, calculate $v_{\mathrm{mA}}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ (v _MB(t), v _MC(t)) and $v_{m0}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain three-phase voltage and residual voltage digital signal v ' _An(v ' _Bn, v ' _Cn), v ' _0nInstantaneous amplitude v _m, v _M0, if residual voltage instantaneous amplitude v _M0Increment v _M0(t+ Δ t)-v _M0(t) greater than residual voltage definite value v _Set4, instantaneous amplitude v _M0(t+ Δ t) is greater than residual voltage definite value v _Set5, and the phase voltage instantaneous amplitude v in the three-phase voltage _m(t+ Δ t) is less than voltage definite value v _Set3, then export the singlephase earth fault confirmation signal.

5), behind the output single-phase earthing confirmation signal, from the residual voltage and the zero-sequence current digital signal v ' of phase-locked sampling module 12 outputs of frequency locking _0n, i ' _0nInsert the sinusoidal approximation process device 52 of ground path judge module 25, calculate  _v(t)=artg (A (t)/B (t)) and  _i(t)=artg (A (t)/B (t)) can obtain residual voltage and zero-sequence current digital signal v ' _0n, i ' _0nInstantaneous phase  _v(t),  _i(t), if the instantaneous phase  of zero-sequence current _i(t) be ahead of the instantaneous phase  of residual voltage _v(t), then export ground path and judge signal, carry out the range finding of singlephase earth fault and calculate.

3, when fault initiating module 2 does not have output short-circuit fault pre-actuation signal or does not export the singlephase earth fault pre-actuation signal, will be from voltage, the current digital signal v ' at the cable two ends that the phase-locked sampling module 12 of frequency locking is exported _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) insert six cable ANN models 6, each cable ANN model 6 inserts a phase voltage and three-phase currents, the single order that calculates input voltage and electric current is to the n order derivative, and with these derivatives also as the input of cable ANN model 6, the output of cable ANN model 6 is voltage v _l, cable ANN model 6 comprises the cable length parameter l, the output v of the cable ANN model 6 at cable two ends _Al ¹(v _Bl ¹, v _Cl ¹) and v _Al ²(v _Bl ², v _Cl ²) difference be the training error e of cable ANN model.Voltage and current input to different changes the cable length parameter, regulates the weight coefficient W of cable ANN model 6 _V1, W _IA1, W _IB1, W _IC1... W _Vn, W _IAn, W _IBn, W _ICn, make the training error e of cable ANN model ²Less than e _Set5, cable ANN model 6 just training finishes.

4, after fault initiating module 2 output short-circuit fault recognition signals or output ground path are judged signal, from voltage, the current digital signal v ' at the cable two ends that the phase-locked sampling module 12 of frequency locking is exported _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) six cable ANN models 6 of access.For short trouble, change the cable length parameter l, make the output v of cable ANN model 6 of the fault phase at cable two ends _l ¹With v _l ²The difference minimum, this moment the cable length parameter l be exactly the distance of trouble spot.For single-phase earthing, change the cable length parameter l, make the output sum of the threephase cable ANN model 6 at cable two ends, i.e. residual voltage v _0l ¹With v _0l ²The difference minimum, this moment the cable length parameter l be exactly the distance of trouble spot.

Claims (9)

1, the online distance-finding method of a kind of cable fault based on artificial nerve network model is characterized in that:

A, the three-phase current simulating signal i that will come from cable two ends current transformer and voltage transformer (VT) _An(i _Bn, i _Cn), zero-sequence current simulating signal i _0nWith three-phase voltage simulating signal v _An(v _Bn, v _Cn), residual voltage simulating signal v _0nInsert low-pass filter circuit (11), electric current, voltage analog signal after low-pass filter circuit (11) is handled are sent into the phase-locked sampling module of frequency locking (12), electric current, the voltage digital signal i ' of output from the phase-locked sampling module of frequency locking (12) _An(i ' _Bn, i ' _Cn), i ' _0n, v ' _An(v ' _Bn, v ' _Cn), v ' _0nBe used for fault initiating module (2), cable ANN (artificial neural network) model training module (3) and cable fault localization computing module (4);

B, fault initiating module (2) are made up of short trouble pretrigger module (21), single-phase earthing pretrigger module (22), short trouble affirmation module (23), single-phase earthing affirmation module (24) and ground path judge module (25), with the current digital signal i ' of output in the phase-locked sampling module of frequency locking (12) _An(i ' _Bn, i ' _Cn) insert the adaptive sine wave filter (51) of short trouble pretrigger module (21), carry out the short trouble pretrigger; The residual voltage digital signal v ' that the phase-locked sampling module of frequency locking (12) is come out _0nInsert the adaptive sine wave filter (51) of single-phase earthing pretrigger module (22), carry out the singlephase earth fault pretrigger;

C, if short trouble pretrigger module (21) output short-circuit fault pre-actuation signal, with voltage, the current digital signal v ' of the phase-locked sampling module of frequency locking (12) output _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) insert the sinusoidal approximation process device (52) that short trouble is confirmed module (23), carry out short trouble and confirm, confirm back output short-circuit fault recognition signal;

D, if single-phase earthing pretrigger module (22) output singlephase earth fault pre-actuation signal, with the three-phase voltage and the residual voltage digital signal v ' of the phase-locked sampling module of frequency locking (12) output _An(v ' _Bn, v ' _Cn), v ' _0nInsert single-phase earthing and confirm the sinusoidal approximation process device (52) of module (24), carry out singlephase earth fault and confirm, confirm back output singlephase earth fault confirmation signal, the residual voltage and the zero-sequence current digital signal v ' of the phase-locked sampling module of frequency locking (12) output _0n, i ' _0nInsert the sinusoidal approximation process device (52) of ground path judge module (25), carry out ground path and judge, the output ground path is judged signal;

After e, fault initiating module (2) output short-circuit fault recognition signal or ground path are judged signal, with voltage, the current digital signal v ' at the cable two ends of the phase-locked sampling module of frequency locking (12) output _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) insert six cable ANN (artificial neural network) models (6), carry out determining of Method of Cable Trouble Point position;

F, if fault initiating module (2) not when output short-circuit fault pre-actuation signal or singlephase earth fault pre-actuation signal, with voltage, the current digital signal v ' at the cable two ends of the phase-locked sampling module of frequency locking (12) output _A1(v ' _B1, v ' _C1), i ' _A1(i ' _B1, i ' _C1) and v ' _A2(v ' _B2, v ' _C2), i ' _A2(i ' _B2, i ' _C2) insert six cable ANN (artificial neural network) models (6), carry out cable ANN model training.

2, according to the online distance-finding method of the described cable fault based on artificial nerve network model of claim 1, it is characterized in that: described adaptive sine wave filter (51) is by one and current digital signal i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches current digital signal i ' _An(i ' _Bn, i ' _Cn) constitute, its mathematical expression is:

y(t)＝Acos(ωt)+Bsin(ωt)

e(t)＝y(t)-i(t)

In order to regulate corrected parameter A and B, carry out following correction algorithm:

A＝A-μe(t)cos(ωt)

B′＝B-μe(t)sin(ωt)

A ', B ' are the modified values of adaptive sine wave filter (51) parameter in the formula, and μ (μ＞0) is the algorithm convergence factor.

3, according to the online distance-finding method of the described cable fault based on artificial nerve network model of claim 1, it is characterized in that: the pretrigger of short trouble described in b method is current digital signal i ' _An(i ' _Bn, i ' _Cn) and the wave filter sinusoidal signal between error e by adaptive sine wave filter (51) output, and with error e and error definite value e _Set1Compare, if e＞e _Set1, and the sum of errors ∑ | e (j) | greater than error definite value e _Set2, output short-circuit fault pre-actuation signal then, on the contrary single-phase earthing pretrigger module (22) then entered.

4, according to the online distance-finding method of the described cable fault based on artificial nerve network model of claim 1, it is characterized in that: the pretrigger of singlephase earth fault described in b method is residual voltage digital signal v ' _0nAnd the error e between the wave filter sinusoidal signal ₀By adaptive sine wave filter (51) output, and with error e ₀With error definite value e _Set3Compare, if e ₀＞e _Set3And sum of errors ∑ | e ₀(j) | greater than error definite value e _Set4, then export the single-phase earthing pre-actuation signal, otherwise then enter cable ANN (artificial neural network) model training (3).

5, according to the online distance-finding method of the described cable fault of claim 1, it is characterized in that based on artificial nerve network model: described sinusoidal approximation process device (52) by one with voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) sinusoidal signal of same frequency approaches voltage, current digital signal v ' in a time domain interval _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) constitute, in sinusoidal approximation process device (52), u (t) is voltage or current digital input signal,

y(t)＝A(t)cosωt+B(t)sinωt

A(t)＝A ₀+A ₁t+A ₂t ²+…+A _Nt ^N

B(t)＝B ₀+B ₁t+B ₂t ²+…+B _Nt ^N

A ₀, A ₁..., A _NAnd B ₀, B ₁..., B _NIt is the parameter of function A (t), B (t).

Note $J=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}e{\left(t_l\right)}^2=\underset{l=1}{\overset{l}{\mathrm{\Σ}}}{(u\left(t_l\right)-y\left(t_l\right))}^2,$ $W^k={\left[\begin{array}{cccccc}{A}_0^k& \·\·\·& A_v^k& B_0^k& \·\·\·& B_v^k\end{array}\right]}^T,$ N is the exponent number of polynomial function, and l is that the calculating in the section is counted, and k is for approaching calculation times.

The correction algorithm of sinusoidal approximation process device (52) is $W^{k+1}=W^k-\mathrm{\μ}\frac{\∂J}{\∂W^k}$ μ (μ＞0) is the algorithm convergence factor.At J＜J _MmThe time, function A (t), B (t) after sinusoidal approximation process device (52) output approaches.

6, according to the online distance-finding method of the described cable fault based on artificial nerve network model of claim 5, it is characterized in that: short circuit fault recognition method is among the c, calculates $v_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ With $i_m\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain voltage, current digital signal v ' _An(v ' _Bn, v ' _Cn), i ' _An(i ' _Bn, i ' _Cn) instantaneous amplitude v _m, i _m, if electric current instantaneous amplitude i _mIncrement i _m(t+ Δ t)-i _m(t) greater than current ration i _Set1, instantaneous amplitude i _m(t+ Δ t) is greater than current ration i _Set2, and voltage instantaneous amplitude v _mIncrement v _m(t)-v _m(t+ Δ t) is greater than voltage definite value v _Set1, instantaneous amplitude v _m(t+ Δ t) is less than voltage definite value v _Set2, output short-circuit fault recognition signal then.

7, according to the online distance-finding method of the described cable fault based on artificial nerve network model of claim 5, it is characterized in that: the singlephase earth fault confirmation method is among the d, calculates $v_{\mathrm{mA}}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}(v_{\mathrm{mB}}\left(t\right),v_{\mathrm{mC}}\left(t\right))$ With $v_{m0}\left(t\right)=\sqrt{A^2\left(t\right)+B^2\left(t\right)}$ Can obtain three-phase voltage and residual voltage digital signal v ' _An(v ' _Bn, v ' _Cn), v ' _0nInstantaneous amplitude v _m, v _M0, if residual voltage instantaneous amplitude v _M0Increment v _M0(t+ Δ t)-v _M0(t) greater than residual voltage definite value v _Set4, instantaneous amplitude v _M0(t+ Δ t) is greater than residual voltage definite value v _Set5, and the phase voltage instantaneous amplitude v in the three-phase voltage _m(t+ Δ t) is less than voltage definite value v _Set3, then export the single-phase earthing confirmation signal; The ground path determination methods is to calculate  _v(t)=artg (A (t)/B (t)) and  _l(t)=artg (A (t)/B (t)) can obtain residual voltage and zero-sequence current digital signal v ' _0n, i ' _0nInstantaneous phase  _v(t),  _l(t), if the instantaneous phase  of zero-sequence current _l(t) be ahead of the instantaneous phase  of residual voltage _v(t), then export ground path and judge signal.

8, according to the online distance-finding method of the described cable fault of claim 1 based on artificial nerve network model, it is characterized in that: the Method of Cable Trouble Point method for determining position is among the e, for short trouble, change the cable length parameter l, make the output vl1 and the v of cable ANN model (6) of the fault phase at cable two ends _l ²The difference minimum, this moment the cable length parameter l be exactly the distance of check point to the trouble spot; For single-phase earthing, change the cable length parameter l, make the output sum of threephase cable ANN (artificial neural network) model (6) at cable two ends, i.e. residual voltage v _0l ¹With v _0l ²The difference minimum, this moment the cable length parameter l be exactly the distance of check point to the trouble spot.

9, according to the online distance-finding method of the described cable fault of claim 1 based on artificial nerve network model, it is characterized in that: cable ANN (artificial neural network) model training method is among the f, each cable ANN model (6) inserts a phase voltage and three-phase current, the single order that calculates input voltage and electric current is to the n order derivative, and with these derivatives also as the input of cable ANN model (6), the output of cable ANN model (6) is voltage v _l, cable ANN model (6) comprises the cable length parameter l, the output v of the cable ANN model (6) at cable two ends _Al ¹(v _Bl ¹, v _Cl ¹) and v _Al ²(v _Bl ², v _Cl ²) difference be the training error e of cable ANN model, to the input of different voltage and current, change the cable length parameter, regulate the weight coefficient W of cable ANN model (6) _Vl, W _IAl, W _IBl, W _ICl... W _Vn, W _IAn, W _IBn, W _ICn, make the training error e of cable ANN model ²Less than e _Set5

Images