Background technology
Electric utility is fast-developing, brings many facilities to daily life, but also brings the electrical fire accident again and again occurred.The electrical fire method and technology of China is delayed, and traditional pick-up unit effectively can not detect arc fault, and when arc fault occurs, moment, localized hyperthermia can arrive 2000 to 4000 degrees Celsius.If do not excise arc fault in time, the serious consequences such as electrical equipment seriously burns, fire, personal injury can be caused.Therefore, be very important work to the detection of fault electric arc.
China starts late in arc-detection technical research, mainly be in the theoretical research stage (number of applying for a patent: 201310376133.4 at present, denomination of invention: the AC fault arc method for measuring based on wavelet transformation and time domain composite character), great majority pass through sensed current signal, analyze current signal waveform and determine whether fault electric arc, but loading condition varies, sometimes break down electric arc time and normal work time the very little (Zou Yunfeng of current waveform difference, Wu Weilin, Li Zhi is brave. based on the low voltage failure electric arc cluster analysis [J] of self-organizing map neural network. and Chinese journal of scientific instrument, 2010, 31 (3) 571-576.), be difficult to determine whether fault electric arc accordingly.
Have comparatively ripe fault electric arc testing product abroad, but voltage environment is different both at home and abroad, arc characteristic can change, and existing procucts can not directly be applied, and therefore, it is very necessary that research is applicable to domestic fault electric arc testing product.
Summary of the invention
For above-mentioned defect of the prior art, a kind of fault arc detection method based on arc spectrum signal that the present invention proposes.
For realizing above-mentioned object, the technical solution used in the present invention is:
The invention provides a kind of fault arc detection method based on arc spectrum signal, first the method gathers spectral signal and current signal by arc spectrum collector, then the normalized based on power is done to spectral signal, and carry out wavelet transformation, using the energy of three first layers wavelet transformation detail signal that the calculates eigenwert as wavelet transformation; Spectral signal is pressed the wavelength coverage disjunction process of the different colours of light, to every section of spectral signal Integral Processing, record the spectral intensity integrated value of every segment; Using the input value of the intensity integrated value of the eigenwert of wavelet transformation and every segment spectrum as BP neural network; Output valve according to BP neural network judges whether to there is fault electric arc.
Specific embodiment of the invention step is as follows:
Step 1: setting spectra collection number of times and the time interval.Spectra collection equipment gathers the arc spectrum signal of electrical contact in a switching process, obtains wavelength and the strength information (λ of spectrum
i, I
i), Hall current sensor gathers current information I simultaneously
a i;
Step 2: judge whether spectra collection number of times reaches setting threshold value, if reach, then starts to perform step 3, otherwise continues step 1;
Step 3: spectral signal is normalized, and carry out wavelet transformation, using the eigenwert of the energy of three first layers wavelet transformation detail signal as wavelet transformation;
Step 4: the wavelength coverage staging treating of spectral signal being pressed the different colours of light, to every section of spectral signal Integral Processing, records the spectral intensity integrated value of every segment;
Step 5: using the input value of the spectral intensity integrated value of the eigenwert of wavelet transformation and every segment as BP neural network;
Step 6: the output valve according to BP neural network judges whether to there is fault electric arc.
Described carries out wavelet transformation to the spectral signal after normalization, obtains the eigenwert of three first layers wavelet transformation details energy, specific as follows: the spectral signal X after normalization carries out wavelet transform, the scale coefficient of the approximate value component that twice filtering obtains.
Described BP neural network, method for building up is as follows:
The details energy eigenvalue of wavelet transformation and the spectral intensity integrated value of different colours are as the input value of BP neural network, MATLAB software is used to set up BP neural network, the transport function of BP neural network from input layer to hidden layer and between hidden layer to output layer adopts logsig and tansig function respectively, and training function adopts traingd function.
Described judges whether arc fault occurs according to the output valve of BP network, is specially: the spectral information training BP neural network repeatedly obtained, refers to when the output valve of neural network is 0 without arc fault, when output valve is 1, arc fault occurs.
Compared with prior art, the invention has the beneficial effects as follows:
The present invention is that a kind of spectral signal that utilizes completely newly carrys out the method for detection failure electric arc, and the size such as wavelength, intensity analyzing spectral signal can obtain burning and produce the material of electric arc, thus understands the using state of electrical contact switch.The present invention has that applicability is wide, security is high, direct reaction electrical contact switch service condition, can accurate failure judgement electric arc.
Embodiment
Be further described technical scheme of the present invention below, the following description is only the use understanding technical solution of the present invention, and be not used in and limit scope of the present invention, protection scope of the present invention is as the criterion with claims.
As shown in Figure 1, method flow diagram for one embodiment of the present invention: a large amount of normal operation first produced by collecting fiber arc generator and the spectral information of fault electric arc, to the spectral signal normalized collected, carry out the eigenwert that wavelet transformation obtains three wavelet transformation details, color segments is pressed and integration to spectrum simultaneously, as the learning sample of BP neural network, by in its input neural network, network is trained, using the recognizer of the neural network trained as fault electric arc.Identification is carried out needing the signal identified to be input in the neural network trained during operation.
As shown in Figure 2, be the fault electric arc recognition device schematic diagram of the present invention one preferred embodiment, wherein: use Ai Wantisi spectrometer, spectral detection scope is 200 ~ 800nm, one point of four optical fiber.When arc generator dynamic/static contact is connected, have electric current to pass through, cam, under the effect of servomotor, drives moving contact motion, makes dynamic/static contact separately, produces electric arc.Photoelectric sensor detects arc light and converts voltage to, and for triggering spectrometer after amplifying shaping, spectrometer receives arc spectrum signal, and utilizes software to process further in the spectral signal computing machine collected.
According to step 1, arrange spectra collection number of times 1000 times, integral time, 1ms, preserved wavelength and the intensity of spectrum.
Step 2: judge whether spectra collection number of times reaches setting threshold value, if reach, then starts to perform step 3:, otherwise continue step 1;
Step 3: spectral signal is normalized, as follows
X
i=I
i/UI
a i
X
ifor the spectral signal after normalization, I
ifor the spectral information before normalization, U is arc voltage, I
a ifor flame current.
Wavelet transformation is carried out to the spectral signal after normalization, obtains the eigenwert of three first layers wavelet transformation details energy, specific as follows: the spectral signal X after normalization carries out wavelet transform, the scale coefficient of the approximate value component that twice filtering obtains.
Calculate three details energy eigenvalues as follows:
d
1=∑W
ψ 2(J-1,k)
d
2=∑W
ψ 2(J-2,k)
I refers to spectra collection number of times, with current acquisition number of times;
W
ψ,
refer to the subspace of spectrum signal function;
H
ψ(l),
refer to scaling function coefficient;
J refers to yardstick, and J-1 is out to out, and smallest dimension is 0;
L refers to the dummy variables in convolution, namely substitutes two relevant sequences;
N=2k, refers to the sequence number of spectral signal, and k refers to the burst of wavelets Subspace;
D
1, d
2, d
3for details energy eigenvalue.
Step 4: the wavelength coverage staging treating of spectral signal being pressed the different colours of light, to every section of spectral signal Integral Processing, records the spectral intensity integrated value of every segment;
Described carries out staging treating by spectral signal by the different colours wavelength of light, and integration, specific as follows:
S
jrefer to the spectral intensity integrated value of jth kind color, b, c are the upper lower limit value of this kind of color spectrum wavelength.
Step 5: using the input value of the spectral intensity integrated value of the eigenwert of wavelet transformation and every segment as BP neural network;
Described BP neural network, method for building up is as follows:
The details energy eigenvalue of wavelet transformation and the spectral intensity integrated value of different colours are as the input value of BP neural network, MATLAB software is used to set up BP neural network, the transport function of BP neural network from input layer to hidden layer and between hidden layer to output layer adopts logsig and tansig function respectively, and training function adopts traingd function.
Step 6: the output valve according to BP neural network judges whether to there is fault electric arc.
Described judges whether arc fault occurs according to the output valve of BP network, is specially: the spectral information training BP neural network repeatedly obtained, refers to when the output valve of neural network is 0 without arc fault, when output valve is 1, arc fault occurs.
The present invention utilizes spectral signal to carry out detection failure electric arc, and the size such as wavelength, intensity analyzing spectral signal can obtain burning and produce the material of electric arc, thus understands the using state of electrical contact switch.The present invention has that applicability is wide, security is high, direct reaction electrical contact switch service condition, can accurate failure judgement electric arc.
The foregoing is only section Example of the present invention, not do any restriction to technical scope of the present invention, all any amendments made within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.