CN103543375A - Method for detecting alternating-current fault arcs on basis of wavelet transformation and time-domain hybrid features - Google Patents

Abstract

The invention relates to a method for detecting alternating-current fault arcs on the basis of wavelet transformation and time-domain hybrid features. The method includes steps of 1), initializing a system; 2), enabling a microprocessor to acquire output current signals of a signal acquisition circuit in real time at set sampling intervals; 3), enabling the microprocessor to judge whether an acquisition number reaches a set threshold value or not, executing a step 4) if the acquisition number reaches the set threshold value, or returning to the step 2) if the acquisition number does not reach the set threshold value; 4), enabling the microprocessor to normalize the sampled current signals in a period, respectively computing time-domain feature values and wavelet transformation feature values of the current signals and inputting the time-domain feature values and the wavelet transformation feature values into a converged BP (back propagation) neural network; 5), judging whether the fault arcs exist or not by the aid of output values of the BP neural network. Compared with the prior art, the method has the advantages of wide application range, high control precision and safety, good expansibility and the like.

Description

AC fault arc method for measuring based on wavelet transformation and time domain composite character

Technical field

The present invention relates to a kind of AC fault arc method for measuring, especially relate to a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character.

Background technology

Along with socioeconomic development, the quantity of fire failure also constantly rises on year-on-year basis, and wherein electrical fire accounts for 30.1% of all fire failure sums.The circuit of house inside, as electric wiring, socket line, household electrical appliance internal wiring and power lead etc., at long-time run with load, transship or affected by external force in the situation that, can cause electric wire insulation layer occur aging or damage in insulation occurs, easily cause the generation of connection in series-parallel arc fault between circuit or line-to-ground short circuit fault.Compare with short trouble, when low voltage failure electric arc occurs, in circuit, current amplitude is less, and traditional pick-up unit cannot detect the electric arc that breaks down, and therefore protection was lost efficacy.The conductive path of the breakdown rear formation of insulating material makes electric arc continue to burn, and the localized hyperthermia producing while burning very easily ignites combustible around and causes fire, serious threat personnel's the security of the lives and property.The protection blind area of traditional detection device becomes fault electric arc to cause the main cause of electrical fire, and the utmost point is necessary that exploitation is for the pick-up unit of low voltage failure electric arc.

The U.S. promotes the use of fault electric arc isolating switch (AFCI the earliest, Arc Fault Circuit Interrupters) country, American insurance business's test has formed UL1699 < < arc fault tripper safety standard > > (Standard for Safety for Arc Fault Circuit Interrupters) in 1999 after further investigation.This standard is defined as AFCI in " a kind of device that can in circuit, fault electric arc be detected and be cut off the electricity supply in time when breaking down electric arc according to arc characteristic ", and it has stipulated performance parameter, safety test test event and the test method etc. of all kinds AFCI.Yet the current difference while producing fault electric arc under different loads is very large, the current characteristic when feature of electric current is with resistive load fault electric arc when some load (as desk lamp with dimmer switch load) normally moves is similar, has brought difficulty to the detection of fault electric arc.So the existing product on its market can only be used in some certain loads situation, there is no good versatility.And the domestic starting of the research for fault electric arc is relatively late, mainly be at present theoretical research stage, from electrology characteristic failure judgement electric arc, the method having proposed comprises that current zero stops time length, current changing rate, electric current maximin, waveform frequency spectrum distribution etc.

Summary of the invention

Object of the present invention be exactly in order to overcome the defect that above-mentioned prior art exists, provide a kind of applied widely, control accuracy is high, the AC fault arc method for measuring based on wavelet transformation and time domain composite character of safe, favorable expandability.

Object of the present invention can be achieved through the following technical solutions:

An AC fault arc method for measuring for wavelet transformation and time domain composite character, is characterized in that, comprises the following steps:

1) system initialization;

2) sampling interval of microprocessor to set, the output current signal of Real-time Collection signal acquisition circuit;

3) whether the number of times of microprocessor judges collection arrives setting threshold, if yes, performs step 4), otherwise return to step 2);

4) current signal of the one-period that microprocessor obtains sampling is normalized, respectively current signal is carried out to time domain and the calculating of Wavelet Transform Feature value, wherein temporal signatures value is the variable quantity of the interior mean value of line current signal half period and peak value, Wavelet Transform Feature value is the energy of front two-layer wavelet transformation detail signal, the input using temporal signatures value and Wavelet Transform Feature value as the BP neural network of convergence;

5) output valve by BP neural network judges whether to exist fault electric arc.

The current signal of the described one-period that sampling is obtained is normalized and is specially:

First by analog-to-digital conversion module, collect array AD _calculate, then through normalized AD (n)=Normal (AD _calculate).

Described temporal signatures value is calculated as follows:

First the current signal AD (n) after normalization is taken absolute value and is obtained | AD (n) |, then calculate respectively the mean value S of its every half period _nwith peak value M _nmean value S with upper half period _n-1with peak value M _n-1, be finally poor obtain temporal signatures value Δ M and Δ S

$\left\{\begin{array}{c}\mathrm{\&Delta;M}=M_n-M_{n-1}\\ \mathrm{\&Delta;S}=S_n-S_{n-1}\end{array}\right.$

。

Described Wavelet Transform Feature value comprises the energy of the detail signal of the two-layer wavelet transformation of current signal, and the detail signal of ground floor wavelet transformation is d ₁(n), its array AD (n) obtaining after by normalization is with wavelet transformation Hi-pass filter coefficient

calculate, concrete formula is:

$d_1\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{g}(2n-k)\mathrm{AD}\left(k\right)$

The approximation signal of ground floor wavelet transformation is a ₁(n), its array AD (n) obtaining after by normalization is with wavelet transformation low-pass filter coefficients

calculate, concrete formula is:

$a_1\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{h}(2n-k)\mathrm{AD}\left(k\right)$

The detail signal of second layer wavelet transformation is d ₂(n), it is by the approximation signal a of ground floor wavelet transformation ₁(n) with wavelet transformation Hi-pass filter coefficient

calculate, concrete formula is:

$d_2\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{g}(2n-k)a_1\left(k\right)$

The last energy e that calculates again the detail signal of two-layer wavelet transformation ₁and e ₂;

? $\left\{\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="e"\; filenames="HDA0000372204530000032.png,HDA0000372204530000041.png"\; state="\{\{state\}\}">e</figure-callout>}}_1=\mathrm{\&Sigma;}{d}_1{\left(n\right)}^2\\ e_2=\mathrm{\&Sigma;}{d}_2{\left(n\right)}^2\end{array}\right.$

。

The BP neural network of described convergence is as follows:

The temporal signatures of the multiple load of group more than recording by experiment and Wavelet Transform Feature value, as the learning sample of neural network, are carried out adaptive learning by MATLAB software to BP neural network, after neural network convergence, are moved in microprocessor.

The described output valve of passing through BP neural network judges whether to exist fault electric arc to be specially:

Network is output as 0 and represents normal condition, output 1 representative has fault electric arc to occur, and judges whether the number of neural network output 1 in setting-up time length is greater than threshold value, if yes, the electric arc that breaks down, microprocessor is controlled tripping mechanism disconnecting circuit by driving circuit.

Described BP neural network input layer is 4 neurons, and output layer is 1 neuron, and the neuronic number in middle layer changes according to the number of sample loadtype.

Described signal acquisition circuit comprises current transformer, voltage follow unit and the signal condition unit connecting successively, and described signal condition unit is connected with microprocessor.

Described fault electric arc detection criteria, can be judged to be fault electric arc while there is eight half cycles electric arc in 0.5 second according to the UL1699 standard code of American insurance business laboratory, and testing circuit should send trip signal; Also can, according to specific requirement, by changing threshold value, adjust.

Compared with prior art, the present invention has the following advantages:

1) applied widely, a kind of detection method and hardware detecting circuit of series fault arc are provided, can realize for the fault electric arc identification of multiple load and the protection of threading off;

2) control accuracy is high, while normally moving for the load that has similar fault electric arc feature in circuit and switch opens or do not produce misoperation when closed.

3) safe, due to the non-linear mapping capability of neural network, can or there is the load of noise pollution correctly to classify by the load of not learning.

4) favorable expandability, for the load newly adding, only needs to adjust the middle layer neuron number of BP neural network, adds the sample of new load and neural network is relearned to convergence.

Accompanying drawing explanation

Fig. 1 is the system construction drawing of fault arc detection device of the present invention;

Fig. 2 is the circuit theory diagrams of fault arc detection device of the present invention;

Fig. 3 is fault electric arc recognition system flow process theory diagram of the present invention;

Fig. 4 is microprocessor program process flow diagram in fault arc detection device of the present invention;

Fig. 5 is typical civilian load---1000W insulating pot fault electric arc oscillogram;

Fig. 6 is typical civilian load---1000W desk lamp with dimmer switch fault electric arc oscillogram;

Fig. 7 is typical civilian load---the low-grade fault electric arc oscillogram of 1000W hair-dryer;

Fig. 8 is typical civilian load---the high-grade fault electric arc oscillogram of 1000W hair-dryer.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.

Embodiment

Fig. 1 and Fig. 2 are respectively overall construction drawing and the circuit theory diagrams of fault arc detection device of the present invention.The time domain based on neural network that the present invention introduces and the fault electric arc recognition device of wavelet transformation composite character as shown in Figure 1, form comprising power module, current collection circuit, microprocessor, driving circuit and releasing structure.Concrete hardware implementation mode is provided by Fig. 2.

Current collection circuit is comprised of current transformer, voltage follow unit and signal condition unit.The current signal of circuit is gathered by current transformer TA1420-05, through sampling resistor R6, the magnitude of current is converted to corresponding voltage signal, the corresponding 5V voltage of specified 25A electric current.Amplifier OP07 and resistance R 7, R8 form voltage follow unit, improved the load capacity of current acquisition signal and sampling resistor and voltage lifting circuit are isolated.STM32 analog to digital conversion port is directly inputted in voltage follow unit, and signal is alternation on the basis of 1.6V DC quantity, can make full use of analog to digital conversion voltage range, improves conversion accuracy.

Microprocessor is mainly comprised of STM32.Its clock frequency reaches as high as 72MHz, and it adopts Harvard structure, has independently instruction bus and data bus, and processor works in pipeline mode, and fetching, decoding and execution can be carried out simultaneously, has improved handling property.Its inner hardware multiplier only needs a clock period just can complete 32 multiply operations, can meet the requirement of Mallat algorithm to computing velocity in wavelet transformation.Major function pin to STM32 arranges as follows:

1) pin 14,21: judgement signal output pin, and distinguish input driving circuit and be connected LED lamp, be set to recommend the way of output.When there is fault electric arc in judgement, pin 14 output high level.LED lamp, for being total to anodic bonding, therefore pin 21 output low levels make LED lamp bright, serves as warning.

2) pin 25: system reset pin, Low level effective, for resetting after electrification reset and failure judgement electric arc.S1 in Fig. 2 is reset button.

3) pin 26: line current signal Gather and input port, in order to analog to digital conversion, is set to analog input mode.

4) pin 35: power frequency square-wave signal input port, in order to trigger external interrupt, reach synchronous effect, and be set to floating empty input mode.

Driving circuit consists of optocoupler P521 and triode Q2.Optocoupler, by microprocessor and isolating switch isolation, plays protection STM32.When driving signal to be high level, the base stage of triode Q2 is high level, and Q2 conducting is also saturated, isolating switch action.The impact of surge voltage when the diode 1N4007 protection triode of Opposite direction connection is avoided isolating switch break-make.

The LY2N-J type isolating switch that tripping mechanism selects Omron Corp to produce.It can disconnect in the 15ms after receiving signal.

Power module selects the fixed frequency off-line power transfer chip VIPer22A that STMicw Electronics produces to form.In electric current and voltage modulate circuit, integrated transporting discharging OP07 needs positive-negative power supply, so the stabilivolt D7 and the D8 that are 7.5V by two voltage stabilizing values are composed in series bleeder circuit, usings its tie point as the earth point of embedded system, can obtain+7.5V voltage source.Shunt capacitance C7 and C8 make output voltage more level and smooth.In addition, use three-terminal voltage-stabilizing chip 7805 and LM1117 respectively general+7.5V voltage transitions be+5.0V and+3.3V, be used as voltage lifting circuit benchmark and be that STM32 powers.

Fig. 3 is fault electric arc recognition system flow process theory diagram of the present invention.System is carried out real-time sampling to the current signal of a large amount of normal operation of unequally loaded and fault electric arc first respectively, sampled result is carried out to eigenwert extraction and as the learning sample of neural network, to in its input neural network, network be trained the recognizer using the neural network training as fault electric arc.During operation, per semiperiod of signal of needs identification is input to and in the neural network training, carries out identification.If recognizer detects over the fault electric arc number of semi-periods of oscillation of number of thresholds and just controls trip gear action at the appointed time, disconnecting circuit, detects otherwise just start a new round.

Fig. 4 is microprocessor program process flow diagram in apparatus of the present invention.As shown in the figure, the workflow of microprocessor is as follows:

1) electrification reset;

2) system initialization, the configuration of I/O mouth, A/D sample mode, sampling rate configuration, system clock, timer configuration;

3) parameter initialization, program parameter assignment;

4) opening timing is interrupted, and starts A/D conversion;

5) with time T _sample=80 μ s are as sampling interval, the output current signal of Real-time Collection signal conditioning circuit;

6) according to the number of times of the signal gathering, whether reach to judge whether duration reaches a power frequency period 0.02 second, is just to proceed for 250 times, the no step 5 that just jumps to.

7) carry out fault electric arc and detect subroutine, the signal of the one-period that sampling obtains to A/D is normalized and obtains AD (n), respectively current signal is carried out the calculating of time domain and Wavelet Transform Feature value.First the signal after normalization is taken absolute value, record peak value M and the mean value S of current half-cycle phase current signal, then calculate the variation delta M of peak value of the same half period and the variation delta S of mean value as two input quantities of neural network; Same normalized signal is carried out to two-layer wavelet decomposition, obtain the detail signal d after two-layer decomposition ₁and d (n) ₂, and calculate their energy value ∑ d (n) ₁(n) ²with ∑ d ₂(n) ²two other input quantity as neural network.Through BP neural computing, obtain Output rusults.

8) whether the number of judgement neural network output 1 within 25 cycles is greater than threshold value 8, if be less than the electric arc that do not break down, jumps to step 5; If be greater than the electric arc that breaks down, continue operation.Microprocessor sends trip signal and drives dropout actuating mechanism disconnecting circuit, until reset button is pressed, circuit is conducting again.

As shown in Fig. 5～8, be respectively 1000W insulating pot fault electric arc oscillogram, 1000W desk lamp with dimmer switch fault electric arc oscillogram, the low-grade fault electric arc oscillogram of 1000W hair-dryer and the high-grade fault electric arc oscillogram of 1000W hair-dryer.Passage 1 is the output waveform of current transformer after signal conditioning circuit, the trip signal of the de-mouthful device of control that passage 2 is microprocessor output, and high level is effective, disconnecting circuit.In Fig. 7, noise filtering is not carried out in the low-grade load of hair-dryer, and other load is all by noise filtering.4 kinds of dissimilar loading on are broken down after electric arc as we can see from the figure, through the identification to the extraction of the collection of current signal, eigenwert and neural network, when pick-up unit detected 8 fault half cycles in 0.5 second, can accurately, timely fault electric arc be detected and be excised.

The present invention is not limited to above embodiment; as long as adopted the time domain based on neural network of the present invention's proposition and the AC fault arc method for measuring of wavelet transformation composite character; other is replaced on an equal basis no matter to adopt which kind of signal conditioning circuit or dropout driving circuit or dropout actuating mechanism or microprocessor etc., within all falling into protection scope of the present invention.

Claims (8)

1. the AC fault arc method for measuring based on wavelet transformation and time domain composite character, is characterized in that, comprises the following steps:

1) system initialization;

2) sampling interval of microprocessor to set, the output current signal of Real-time Collection signal acquisition circuit;

3) whether the number of times of microprocessor judges collection arrives setting threshold, if yes, performs step 4), otherwise return to step 2);

4) current signal of the one-period that microprocessor obtains sampling is normalized, respectively current signal is carried out to time domain and the calculating of Wavelet Transform Feature value, wherein temporal signatures value is the variable quantity of the interior mean value of line current signal half period and peak value, Wavelet Transform Feature value is the energy of front two-layer wavelet transformation detail signal, the input using temporal signatures value and Wavelet Transform Feature value as the BP neural network of convergence;

5) output valve by BP neural network judges whether to exist fault electric arc.

2. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 1, is characterized in that, the current signal of the described one-period that sampling is obtained is normalized and is specially:

First by analog-to-digital conversion module, collect array AD _calculate, then through normalized AD (n)=Normal (AD _calculate).

3. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 2, is characterized in that, described temporal signatures value is calculated as follows:

First the current signal AD (n) after normalization is taken absolute value and is obtained | AD (n) |, then calculate respectively the mean value S of its every half period _nwith peak value M _nmean value S with upper half period _n-1with peak value M _n-1, be finally poor obtain temporal signatures value Δ M and Δ S

$\left\{\begin{array}{c}\mathrm{\&Delta;M}=M_n-M_{n-1}\\ \mathrm{\&Delta;S}=S_n-S_{n-1}\end{array}\right.$

。

4. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 2, it is characterized in that, described Wavelet Transform Feature value comprises the energy of the detail signal of the two-layer wavelet transformation of current signal, and the detail signal of ground floor wavelet transformation is d ₁(n), its array AD (n) obtaining after by normalization is with wavelet transformation Hi-pass filter coefficient

calculate, concrete formula is:

$d_1\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{g}(2n-k)\mathrm{AD}\left(k\right)$

The approximation signal of ground floor wavelet transformation is a ₁(n), its array AD (n) obtaining after by normalization is with wavelet transformation low-pass filter coefficients

calculate, concrete formula is:

$a_1\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{h}(2n-k)\mathrm{AD}\left(k\right)$

The detail signal of second layer wavelet transformation is d ₂(n), it is by the approximation signal a of ground floor wavelet transformation ₁(n) with wavelet transformation Hi-pass filter coefficient

calculate, concrete formula is:

$d_2\left(n\right)=\underset{k}{\mathrm{\&Sigma;}}\stackrel{\&OverBar;}{g}(2n-k)a_1\left(k\right)$

The last energy e that calculates again the detail signal of two-layer wavelet transformation ₁and e ₂;

? $\left\{\begin{array}{c}{e}_1=\mathrm{\&Sigma;}{d}_1{\left(n\right)}^2\\ e_2=\mathrm{\&Sigma;}{d}_2{\left(n\right)}^2\end{array}\right.$

。

5. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 1, is characterized in that, the BP neural network of described convergence is as follows:

The temporal signatures of the multiple load of group more than recording by experiment and Wavelet Transform Feature value, as the learning sample of neural network, are carried out adaptive learning by MATLAB software to BP neural network, after neural network convergence, are moved in microprocessor.

6. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 1, is characterized in that, the described output valve of passing through BP neural network judges whether to exist fault electric arc to be specially:

Network is output as 0 and represents normal condition, output 1 representative has fault electric arc to occur, and judges whether the number of neural network output 1 in setting-up time length is greater than threshold value, if yes, the electric arc that breaks down, microprocessor is controlled tripping mechanism disconnecting circuit by driving circuit.

7. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 1, it is characterized in that, described BP neural network input layer is 4 neurons, output layer is 1 neuron, and the neuronic number in middle layer changes according to the number of sample loadtype.

8. a kind of AC fault arc method for measuring based on wavelet transformation and time domain composite character according to claim 1, it is characterized in that, described signal acquisition circuit comprises current transformer, voltage follow unit and the signal condition unit connecting successively, and described signal condition unit is connected with microprocessor.

Images