CN101334428A - Method and device for separating voltage electric current signal of electrical power system - Google Patents

Abstract

The invention discloses a method and a device for separating voltage and current signals of a power system, belonging to the technical field of signal separation; the device consists of a lower computer and an upper computer, the lower computer comprises a sensor, a signal conditioning module, an A/D converting module, a dual single chip system unit, a storage module, a keyboard module and a communication module, wherein, the sensor, the signal conditioning module, the A/D converting module and the dual single chip system unit are sequentially connected, the storage module, the keyboard operation unit module and the communication module are respectively connected with the dual signal chip system unit, and the upper computer is connected with the communication module. The method is completed by software part, the software part is embedded into the dual single chip system unit, and a separation matrix is multiplied by a source signal to obtain a final separation signal expression formula. The method and the device can reduce the storage space and improve the precision of records, thereby approximating the actual waveform by arbitrary precision within any small time, well analyzing the composition of harmonic wave in signals and providing data basis for harmonic analysis.

Description

A kind of method and device that the power system voltage current signal is separated

Technical field

The invention belongs to the signal separation techniques field, particularly a kind of method and device that the power system voltage current signal is separated.

Background technology

Harmonic wave roughly has the following aspects to the harm of electrical network: 1. harmonic wave makes the element in the electrical network produce additional harmonic loss, reduces the efficient of generating, transmission of electricity and consumer, and a large amount of 3 subharmonic can make the overheated even breaking out of fire of circuit when flowing through the neutral line; 2. the operate as normal of the various electrical equipments of harmonic effects makes motor produce mechanical vibration, noise and superpotential, makes the transformer part seriously overheated, makes apparatus overheat, insulation ag(e)ing, the losts of life such as electric capacity, cable, so that damage; 3. harmonic wave can cause parallel resonance and series resonance local in the electrical network, thereby harmonic wave is amplified, and causes major accident; 4. harmonic wave can cause the misoperation of relay protection and aut.eq., makes the electrical measuring instrument metering inaccurate; 5. harmonic wave can produce contiguous communication system and disturb, and the lighter produces noise, reduces communication quality; Weight person causes information dropout, makes the communication system can't operate as normal.A kind of technology of isolating source signal from the mixing of a plurality of independent source signals that is meant is separated in blind source.The blind source separation algorithm of multiple effective instantaneous mixing has been proposed in the world, originally algorithm source signal only superelevation this or be could obtain preferable performance under the inferior gaussian signal situation entirely, otherwise, be difficult to reach ideal effect. at the mixed signal separation problem of the source signal that has multiple probability distribution simultaneously, scholars have proposed improving one's methods of some.Based on the parallel separation algorithm that nonlinear function is selected, the selection of its function is relevant with the probability density function of source signal, need constantly switch nonlinear function in estimation procedure; Serial algorithm need carry out orthonomalization or go redundant the processing. so these two kinds of algorithms all need to carry out iterative computation, and algorithm complexity and calculated amount are bigger.

Summary of the invention

At the prior art weak point, the invention provides a kind of method and device that the power system voltage current signal is separated.

The technical solution used in the present invention is: combining with distribution system based on the algorithm of PID neural network with based on the algorithm of maximum signal to noise ratio during blind source data is separated, consider the actual conditions of system, the voltage and current signal that collects separated and provide it embody formula, having solved can only the bigger signal of filtering interfering at signal filter circuit, and to being blended in the effectively situation of filtering of signal that amplitude in the useful signal and frequency be more or less the same.And in device, adopted dual-CPU system, system to be divided into the first and second two chips, and the first machine is responsible for the collection of signal, and the second machine is responsible for the separating treatment of signal, improves the work efficiency of system greatly.

Device of the present invention is made up of slave computer and host computer, slave computer comprises sensor, signal condition module, A/D modular converter (analog-to-digital conversion module just), double mcu system unit, memory module, Keysheet module, communication module, wherein sensor, signal condition module, A/D modular converter, double mcu system unit connect successively, memory module, keyboard operation unit module, communication module link to each other with the double mcu system unit respectively, and host computer links to each other with communication module.

Wherein sensor comprises the voltage transformer (VT) summation current transformer, the voltage and current of acquisition system, be input to the signal condition module, the signal condition module is regulated its size by proper proportion and is made it to become the treatable amplitude range of device, the electric current and voltage of A/D modular converter acknowledge(ment) signal conditioning module conditioning is converted into the digital signal that the double mcu system unit can be discerned; The double mcu system unit comprises first machine and second machine, and the first machine is responsible for the control data gatherer process, and the second machine is according to certain algorithm principle separation signal, and the first machine links to each other by conversion chip with the second machine; The first machine is connected with memory module by conversion chip 74ALS373, and the second machine directly links to each other with memory module; Keysheet module is connected with the double mcu system unit by parallel interface, is used for the various correlation parameters of input system.The identification result that communication module is handled the double mcu system unit is sent to host computer, shows correlation parameter and Signal Separation image at last in the host computer liquid crystal display, and the serial line interface of communication module adopts the RS232 agreement.

Method of the present invention is finished by software section, and software section embeds the double mcu system unit, may further comprise the steps:

1, beginning

2, gather source signal

Use device is gathered voltage, the electric current of electric system, is transferred to the double mcu system unit.It specifically is voltage and current by the sensor acquisition electric system, be input to the signal condition module, the signal condition module is regulated its size by proper proportion and is made it to become the treatable amplitude range of device, the electric current and voltage of A/D modular converter acknowledge(ment) signal conditioning module conditioning is converted into the digital signal that the double mcu system unit can be discerned.

3, determine the separated network structure

After collecting signal, separate according to the network structure that the number of the aliasing signal that will separate is selected, for example there are 6 road signals in system, it is corresponding with the input source signal to construct 6 sub-PID neuroids, network structure promptly has 6 sub-PID neuroids, the structure that can set each sub-network arbitrarily is 2-2-2 or 3-3-3, the texture ratio 2-2-2 structure precision height of 3-3-3, single PI neural network is 3 layers of feedforward network, its structure is 2-2-2, the input layer of network is made of two neurons, the blind signal of aliasing of corresponding two inputs; Hidden layer also has two neurons, and each neuronic output function is different, corresponds respectively to ratio (P), two parts of integration (I); The output layer of network is exactly the signal of two separation.The forward calculation of network realizes the calculation task of decomposition, and the inverse algorithms of network is realized the self-adaptation adjustment of pid parameter.

Arthmetic statement is as follows:

$v_j=\underset{m=1}{\overset{2}{\mathrm{\Σ}}}{w}_{\mathrm{jm}}^2{x}_m,$ j＝1，2，m＝1，2

In the formula: x _mBe m input source signal; w _Jm ²Be the weights from the m input layer to the j hidden layer, 2 are from the input layer to the hidden layer; v _jIt is the input value of j hidden layer.Order $w_{}$ ${}_2=\left(w_{\mathrm{jm}}^2\right),$ () expression is by w _Jm ²The matrix of forming, down together.

To the hidden layer of PI neural network, it has comprised ratio unit and integration unit.Ratio unit state is

$u_1\left(k\right)=\left\{\begin{array}{cc}10& u_1\left(k\right)>10\\ u_1\left(k\right)& -10\&le;u_1\left(k\right)\&le;10\\ -10& u_1\left(k\right)<-10\end{array}\right.$

Integration unit state is

$u_2\left(k\right)=\left\{\begin{array}{cc}100& u_2\left(k\right)>100\\ u_2\left(k\right)+v_2\left(k\right)& -100\&le;u_2\left(k\right)\&le;100\\ -100& u_2\left(k\right)<-100\end{array}\right.$

Output layer is neuronic to be output as

$y_n=\underset{n=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{nj}}^1{u}_j,$ j＝1，2，n＝1，2

In the formula: u _jIt is the output valve of j hidden layer; w _Nj ¹Be the weights from j hidden layer to n output layer, 1 is from the hidden layer to the output layer; y _nBe n output valve.Order $W_{}$ ${}_1={\mathrm{<figure-callout\; id="1"\; label="w\; nj"\; filenames="A20081001141900162.png,A20081001141900171.png"\; state="\{\{state\}\}">w</figure-callout>}}_{\mathrm{nj}}^{}.$

It is y=W that then whole PI network is abbreviated as matrix form ₁G (u (w ₂X))

Single PID neural network structure is 3-3-3.Its hidden layer has three neurons, corresponds respectively to ratio (P), integration (I) and three parts of differential (D).Contrast PI neural network, hidden layer has only increased a differential (D), and its state expression formula is

$u_3\left(k\right)=\left\{\begin{array}{cc}10& u_3\left(k\right)>10\\ u_3\left(k\right)-v_3(k-1)& -10\&le;u_3\left(k\right)\&le;10\\ -10& u_3\left(k\right)<-10\end{array}\right.$

When the input of more signal, it is corresponding with the input source signal to construct a plurality of sub-PID neuroids.

4, set non-linear aliasing model matrix and isolate the estimated signal of signal as source signal

By element w _Jm ², w _Nj ¹The matrix of forming is called non-linear aliasing model matrix, and the estimated value of source signal is y=W ₁G (u (w ₂X)).

Because original signal is respectively from different signal sources, therefore think between each source signal it is separate, promptly there be n from the statistics of signal source signal independently, s _i(t), the mixed signal x of i=1～n and similar number _i(t), the mixing of signal is a linear instantaneous, and the mixed signal that observes is

$x_i\left(t\right)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_{\mathrm{ij}}{s}_j\left(t\right),$ A in the formula _IjBe mixing constant.Blind source separation problem is exactly only to be supposed by the statistical independence between the x that observes (t) and each component of s (t), and recovers s (t) by some priori to the probability distribution of original defeated people's signal.Problem description can be become the separation matrix W on nxn rank, it is output as

y(t)＝Wx(t)＝WAs(t)＝Gs(t)

Overall situation matrix G=PD (P is a nxn dimension permutation matrix, and D is a nxn dimension diagonal matrix), then y (t)=PDs (t).Signal y (t) after the recovery compares with source signal s (t), changes to some extent on amplitude and order, and it is uncertain that this is called the inherence of separating in blind source.Because information spinner will be included in the waveform of signal, so these two kinds of uncertainties do not influence the application of blind source separate technology.At first be that to set up one be the objective function F (W) of argument with W, if certain

Can make F (W) reach greatly (little) value, should

Be required separating; Objective function of the present invention is the signal to noise ratio (S/N ratio) function, and optimizing process has caused generalized eigenvalue problem, therefore just can obtain separation matrix under the situation of iteration

5, calculate the blind source of the error foundation disjunctive model of source signal and its estimated signal, error e=s-y of source signal s and its estimated signal y as noise signal;

6, set up the signal to noise ratio (S/N ratio) function of restoring signal

Set up the signal to noise ratio (S/N ratio) function of restoring signal s: $\mathrm{SNR}=10\log \frac{s*s^T}{e*e^T}=\frac{s*s^T}{(s-y)*{(s-y)}^T}.$

7, set up the maximum signal to noise ratio objective function

Because source signal is unknown, it is as follows therefore to obtain the maximum signal to noise ratio objective function with the running mean y replacement source signal of estimated signal: $F\left(y\right)=\mathrm{SNR}=10\log \frac{y*y^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}.$

Y=Wx in the formula, y=Wx, x are the signal of mixed signal x after running mean is handled.

8, optimization aim function

The maximum signal to noise ratio objective function can be write as following form:

$F(W,x)=10\log \frac{y{*y}^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}$

$=10\log \frac{\mathrm{Wx}*x^T{W}^T}{(s-y)*{(s-y)}^T}$

$=10\log \frac{{\mathrm{WCW}}^T}{W\stackrel{\&OverBar;}{C}{W}^T}$

$=10\log \frac{V}{U}$

9, calculate the separation matrix gradient

The gradient of the function F of hence one can see that separation matrix W is $\frac{\&PartialD;F}{\&PartialD;W}=\frac{2W}{V}C-\frac{2W}{U}\stackrel{\&OverBar;}{C},$ Can try to achieve separation matrix thus

10, separation matrix and source signal multiply each other and obtain final separation signal expression formula

Separation matrix and source signal multiply each other and obtain final separation signal expression formula.

11, the first-harmonic of voltage and current signal and the host computer that is delivered to of harmonic wave thereof are also shown.

The invention has the beneficial effects as follows: the separation method that the mode that adopts original electric system recording voltage current signal the pointwise record to form waveform has been modified into based on the blind source data of neuroid and maximum signal to noise ratio obtains the first-harmonic of voltage and current signal and the stack expression formula of harmonic wave thereof, this form has realized that the vector quantization record of waveform has effectively reduced the precision of the record of storage space and raising, can in the arbitrarily small time, approach actual waveform, to the waveform inversion of electric network fault, recollect accurate record is provided with arbitrary accuracy.By the mode of expression formula record, can well analytic signal in the composition of harmonic wave, for frequency analysis provides the data basis.

Description of drawings

Fig. 1 structure drawing of device of the present invention;

Fig. 2 signal condition module circuit diagram of the present invention;

Fig. 3 A/D modular converter of the present invention circuit diagram;

Fig. 4 double mcu system unit of the present invention and memory module circuit diagram;

Fig. 5 Keysheet module circuit diagram of the present invention;

Fig. 6 communication module circuit diagram of the present invention;

Fig. 7 software flow pattern of the present invention;

Fig. 8 a kind of single PID neural network synoptic diagram of the present invention;

Fig. 9 another single PID neural network synoptic diagram of the present invention;

The output map as a result of Figure 10 a kind of embodiment of the present invention.

Embodiment

In conjunction with the accompanying drawings the present invention is described further:

As shown in Figure 1, apparatus module figure of the present invention, form by slave computer and host computer, slave computer comprises that sensor, signal condition module, A/D modular converter are analog-to-digital conversion module, double mcu system unit, memory module, Keysheet module, communication module, wherein sensor, signal condition module, A/D modular converter, double mcu system unit connect successively, memory module, keyboard operation unit module, communication module link to each other with the double mcu system unit respectively, and host computer links to each other with communication module.

Wherein sensor comprises the voltage transformer (VT) summation current transformer, the voltage and current of acquisition system, be input to the signal condition module, the signal condition module is regulated its size by proper proportion and is made it to become the treatable amplitude range of device, the electric current and voltage of A/D modular converter acknowledge(ment) signal conditioning module conditioning is converted into the digital signal that the double mcu system unit can be discerned; The double mcu system unit comprises first machine and second machine, and the first machine is responsible for the control data gatherer process, and the second machine is according to certain algorithm principle separation signal, and the first machine links to each other by conversion chip with the second machine; The first machine is connected with memory module by chip 74ALS373, and the second machine directly links to each other with storage chip; Keysheet module is connected with the double mcu system unit by parallel interface, is used for the various correlation parameters of input system.The identification result that communication module is handled the double mcu system unit is sent to host computer, shows correlation parameter and Signal Separation image at last in the host computer liquid crystal display, and the serial line interface of communication module adopts the RS232 agreement.

Describe of the present invention in conjunction with one embodiment of the present of invention, voltage transformer (VT) is selected PT204A for use, and current transformer is selected CT254A for use.

As shown in Figure 2, the circuit diagram of signal condition module, described voltage and current signal condition will be sent to amplifier LM258P through the signal after superpotential, the current transformer conversion and amplify, nurse one's health-5V～+ voltage of 5V scope is input to analog-to-digital conversion module and samples.

As shown in Figure 3, A/D modular converter circuit diagram, described A/D modular converter is selected chip MAX125 for use, the output signal of the channel receiving signal conditioning module of pin 1,2,3,4,33,34, three-phase voltage electric current to the output of signal condition module is sampled, and the master data of system-computed is provided.P1.0～P1.7 the pin of the first machine of the output pin D0 of A/D conversion chip～D7 while and double mcu system, D0～D7 pin of conversion chip 74ALS373 link to each other, and the ALE port of first machine links to each other with 11 pins of conversion chip 74ALS373.

As shown in Figure 4, double mcu system unit and memory module circuit diagram, first machine and second machine adopt the 87C51 chip, and the first machine is responsible for the collection of signal, and the second machine is responsible for the separating treatment of signal, improves the work efficiency of system greatly.First machine and second machine adopt the 87C51 chip, and the first machine is responsible for the collection of signal, and the second machine is responsible for the separating treatment of signal, improves the work efficiency of system greatly.First machine P1.0 mouth links to each other with second machine INT0 mouth, and second machine P1.7 mouth links to each other with first machine INT0 mouth.First machine address wire links to each other with Q0～Q7 delivery outlet of chip 74ALS373.Memory module selects for use the D0～D7 port of chip 6264,6264 chips and B0～B7 delivery outlet of chip 74ALS245 to link to each other, and Q0～Q7 delivery outlet of the A0 of 6264 chips～A7 port and chip 74ALS373 links to each other.The first machine is connected with storage chip by chip 74AL373, and the second machine directly links to each other with storage chip.Storage chip except the parameters that storage system is provided with, also store 64 days on schedule the time each phase voltage, electric current, the reserve battery of storer guarantees that data under the situation of system's electricity shortage, do not lose.

As shown in Figure 5, the Keysheet module circuit diagram, described keyboard operation module is used for some man-machine conversational operations, is provided with etc. such as the parameter that needs the people to device, and the content of this part is finished by button is set in the hardware design.This module plays the conversion effect between PC keyboard and single-chip microcomputer.It has shielded the function of carrying out the complex process of data and command interaction with the PC keyboard and having realized keyboard interrupt service routine in the similar dos operating system, makes the designer only need be concerned about the result who receives button, and can use the keyboard coding of standard to programme.It requires single-chip microcomputer to join with it by 8 bit parallel interfaces, directly is connected with the P1.0～P1.7 mouth of first machine by plug connector, and keyboard and main frame join by keyboard plug, and keyboard plug has two kinds of the big plug of 5 cores and 6 core small plugs (PS/2 interface).Interface signal has: power supply,, keyboard clock KB CLK, keyboard data KB DAT.During operate as normal, keyboard circuit is the keyboard scan matrix constantly.If there is key to press, then send the position scan code of button to the mainboard keyboard interface circuit with serial mode.When pressing key, send and connect scan code, during reverse keys, send the disconnection scan code of this key.Disconnecting scan code generally is to add one to disconnect flag byte F0H before connecting scan code.If certain key is pressed always, then send the connection scan code of this key continuously with the button repetition rate.Scan code is relevant with the position of button, there is no corresponding relation with the ASCII character of this key.QWERTY keyboard is two-way with communicating by letter of main frame, and adopts 11 serial asynchronous communication form, and this 11 bit data comprises: start bit 0,8 bit data positions (LSB formerly), odd parity bit P, position of rest 1.

As shown in Figure 6, the communication module circuit diagram, communication module has adopted the MAX232 chip, and described communication module is considered the needs at comprehensive automation station, so increased the function that communicates contact with upper level department.Serial port itself only provides hardware configuration and basic communication work mode to mutual channel.The computing machine serial line interface adopts the RS232 agreement.The pin 11,12 of MAX232 links to each other with single-chip microcomputer first pin of chip 10,11.232 sockets link to each other with the host computer serial ports by 232 cables.

As shown in Figure 7, software flow pattern of the present invention,

1, beginning

2, gather source signal

Use device is gathered voltage, the electric current of electric system, is transferred to the double mcu system unit.

3, determine the separated network structure

We can separate according to the network structure that the number of the aliasing signal that will separate is selected after collecting signal, for example there are 6 road signals in system, it is corresponding with the input source signal to construct 6 sub-PID neuroids, network structure promptly has 6 sub-PID neuroids, the structure that can set each sub-network arbitrarily is 2-2-2 or 3-3-3, the texture ratio 2-2-2 structure precision height of 3-3-3, single PI neural network is 3 layers of feedforward network, its structure is 2-2-2, the input layer of network is made of two neurons, the blind signal of aliasing of corresponding two inputs; Hidden layer also has two neurons, and each neuronic output function is different, corresponds respectively to ratio (P), two parts of integration (I); The output layer of network is exactly the signal of two separation.The forward calculation of network realizes the calculation task of decomposition, and the inverse algorithms of network is realized the self-adaptation adjustment of pid parameter.

Arthmetic statement is as follows:

$v_j=\underset{m=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{jm}}^2{x}_m,$ j＝1，2，m＝1，2

In the formula: x _mBe m input source signal; w _Jm ²Be the weights from the m input layer to the j hidden layer, 2 are from the input layer to the hidden layer; v _jIt is the input value of j hidden layer.Order $w_{}$ ${}_2=\left(w_{\mathrm{jm}}^2\right),$ () expression is by w _Jm ²The matrix of forming, down together.

To the hidden layer of PI neural network, it has comprised ratio unit and integration unit.Ratio unit state is

$u_1\left(k\right)=\left\{\begin{array}{cc}10& u_1\left(k\right)>10\\ u_1\left(k\right)& -10\&le;u_1\left(k\right)\&le;10\\ -10& u_1\left(k\right)<-10\end{array}\right.$

Integration unit state is

$u_2\left(k\right)=\left\{\begin{array}{cc}100& u_2\left(k\right)>100\\ u_2\left(k\right)+v_2\left(k\right)& -100\&le;u_2\left(k\right)\&le;100\\ -100& u_2\left(k\right)<-100\end{array}\right.$

Output layer is neuronic to be output as

$y_n=\underset{n=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{nj}}^1{u}_j,$ j＝1，2，n＝1，2

In the formula: u _jIt is the output valve of j hidden layer; w _Nj ¹Be the weights from j hidden layer to n output layer, 1 is from the hidden layer to the output layer; y _nBe n output valve.Order $W_{}$ ${}_1=w_{\mathrm{nj}}^1.$

Then whole PI network is abbreviated as matrix form

y＝W ₁g(u(w ₂x))

Single PID neural network structure is 3-3-3.Its hidden layer has three neurons, corresponds respectively to ratio (P), integration (I) and three parts of differential (D).Contrast PI neural network, hidden layer has only increased a differential (D), and its state expression formula is

$u_3\left(k\right)=\left\{\begin{array}{cc}10& u_3\left(k\right)>10\\ u_3\left(k\right)-v_3(k-1)& -10\&le;u_3\left(k\right)\&le;10\\ -10& u_3\left(k\right)<-10\end{array}\right.$

When the input of more signal, it is corresponding with the input source signal to construct a plurality of sub-PID neuroids.

4, set non-linear aliasing model matrix and isolate the estimated signal of signal as source signal

By element w _Jm ², w _Nj ¹The matrix of forming is called non-linear aliasing model matrix, and the estimated value of source signal is y=W ₁G (u (w ₂X)).

Because original signal is respectively from different signal sources, therefore think between each source signal it is separate, promptly there be n from the statistics of signal source signal independently, s _i(t), the mixed signal x of i=1～n and similar number _i(t), the mixing of signal is a linear instantaneous, and the mixed signal that observes is

$x_i\left(t\right)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_{\mathrm{ij}}{s}_j\left(t\right),$ A in the formula _IjBe mixing constant.Blind source separation problem is exactly only to be supposed by the statistical independence between the x that observes (t) and each component of s (t), and recovers s (t) by some priori to the probability distribution of original defeated people's signal.Problem description can be become the separation matrix W on nxn rank, it is output as

y(t)＝Wx(t)＝WAs(t)＝Gs(t)

Overall situation matrix G=PD (P is a nxn dimension permutation matrix, and D is a nxn dimension diagonal matrix), then y (t)=PDs (t).Signal y (t) after the recovery compares with source signal s (t), changes to some extent on amplitude and order, and it is uncertain that this is called the inherence of separating in blind source.Because information spinner will be included in the waveform of signal, so these two kinds of uncertainties do not influence the application of blind source separate technology.At first be that to set up one be the objective function F (W) of argument with W, if certain

Can make F (W) reach greatly (little) value, should

Be required separating; Objective function of the present invention is the signal to noise ratio (S/N ratio) function, and optimizing process has caused generalized eigenvalue problem, therefore just can obtain separation matrix under the situation of iteration

5, calculate the error of source signal and its estimated signal

According to blind source disjunctive model, error e=s-y of source signal s and its estimated signal y as noise signal;

6, set up the signal to noise ratio (S/N ratio) function of restoring signal

Set up the signal to noise ratio (S/N ratio) function of restoring signal s: $\mathrm{SNR}=10\log \frac{s*s^T}{e*e^T}=\frac{s*s^T}{(s-y)*{(s-y)}^T}.$

7, set up the maximum signal to noise ratio objective function

Because source signal is unknown, it is as follows therefore to obtain the maximum signal to noise ratio objective function with the running mean y replacement source signal of estimated signal: $F\left(y\right)=\mathrm{SNR}=10\log \frac{y*y^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}.$

Y=Wx in the formula, y=Wx, x are the signal of mixed signal x after running mean is handled.

8, optimization aim function

The maximum signal to noise ratio objective function can be write as following form:

$F(W,x)=10\log \frac{y{*y}^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}$

$=10\log \frac{\mathrm{Wx}*x^T{W}^T}{(s-y)*{(s-y)}^T}$

$=10\log \frac{{\mathrm{WCW}}^T}{W\stackrel{\&OverBar;}{C}{W}^T}$

$=10\log \frac{V}{U}$

9, calculate the separation matrix gradient

The gradient of the function F of hence one can see that separation matrix W is $\frac{\&PartialD;F}{\&PartialD;W}=\frac{2W}{V}C-\frac{2W}{U}\stackrel{\&OverBar;}{C},$ Can try to achieve separation matrix thus

10, separation matrix and source signal multiply each other and obtain final separation signal expression formula

Separation matrix and source signal multiply each other and obtain final separation signal expression formula.

11, the first-harmonic of voltage and current signal and the host computer that is delivered to of harmonic wave thereof are also shown.

Shown in Fig. 8 and 9, single PID neural network synoptic diagram of the present invention, single PI neural network shown in Figure 8 is 3 layers of feedforward network, and its structure is 2-2-2, and the input layer of network is made of two neurons, the blind signal of aliasing of corresponding two inputs; Hidden layer also has two neurons, and each neuronic output function is different, corresponds respectively to ratio (P), two parts of integration (I); The output layer of network is exactly the signal of two separation.Arthmetic statement is as follows:

$v_j=\underset{m=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{jm}}^2{x}_m,$ j＝1，2，m＝1，2

In the formula: x _mBe m input source signal; w _Jm ²Be the weights from the m input layer to the j hidden layer, 2 are from the input layer to the hidden layer; v _jIt is the input value of j hidden layer.Order $w_{}$ ${}_2=\left(w_{\mathrm{jm}}^2\right),$ () expression is by w _Jm ²The matrix of forming, down together.

To the hidden layer of PI neural network, it has comprised ratio unit and integration unit.Ratio unit state is

$u_1\left(k\right)=\left\{\begin{array}{cc}10& u_1\left(k\right)>10\\ u_1\left(k\right)& -10\&le;u_1\left(k\right)\&le;10\\ -10& u_1\left(k\right)<-10\end{array}\right.$

Integration unit state is

$u_2\left(k\right)=\left\{\begin{array}{cc}100& u_2\left(k\right)>100\\ u_2\left(k\right)+v_2\left(k\right)& -100\&le;u_2\left(k\right)\&le;100\\ -100& u_2\left(k\right)<-100\end{array}\right.$

Output layer is neuronic to be output as

$y_n=\underset{n=1}{\overset{2}{\mathrm{\&Sigma;}}}{w}_{\mathrm{nj}}^1{u}_j,$ j＝1，2，n＝1，2

In the formula: u _jIt is the output valve of j hidden layer; w _Nj ¹Be the weights from j hidden layer to n output layer, 1 is from the hidden layer to the output layer; y _nBe n output valve.Order $W_{}$ ${}_1=w_{\mathrm{nj}}^1.$

Then whole PI network is abbreviated as matrix form

y＝W ₁g(u(w ₂x))

Single PID neural network structure shown in Figure 9 is 3-3-3.Its hidden layer has three neurons, corresponds respectively to ratio (P), integration (I) and three parts of differential (D).Contrast PI neural network, hidden layer has only increased a differential (D), and its state expression formula is

$u_3\left(k\right)=\left\{\begin{array}{cc}10& u_3\left(k\right)>10\\ u_3\left(k\right)-v_3(k-1)& -10\&le;u_3\left(k\right)\&le;10\\ -10& u_3\left(k\right)<-10\end{array}\right.$

Electric current with the mountain 66KV primary station collection of Tongliao Baolong is an example now, carries out harmonic separation.

When system acquisition behind current signal, be sent to amplifier LM258P by current transformer CT254A and amplify, nurse one's health-5V～+ voltage of 5V scope is input to analog-to-digital conversion module and samples.The A/D modular converter is selected chip MAX125 for use, and the output signal of the channel receiving signal conditioning module of pin 1,2,3,4,33,34 is sampled to the three-phase current of signal condition module output, and the master data of system-computed is provided.The A/D modular converter is sent to the first machine to the result of sampling, double mcu system coordination work, the first machine is responsible for the collection of signal, and the second machine is responsible for the separating treatment of signal, the first machine is sent out " a 1 " negative edge pulse for the INT0 end of second single-chip microcomputer by P1.0 and is waken the second machine up, allow the second machine continue working procedure.The second machine is in the result of calculation memory module.And adopt the RS232 agreement that the result is presented in the host computer by the computing machine serial line interface.

Adopting neural network structure in this example is 2-2-2.To the hidden layer of PI neural network, it has comprised ratio unit and integration unit.Ratio unit state is

$u_1\left(k\right)=\left\{\begin{array}{cc}10& u_1\left(k\right)>10\\ u_1\left(k\right)& -10\&le;u_1\left(k\right)\&le;10\\ -10& u_1\left(k\right)<-10\end{array}\right.$

Integration unit state is

$u_2\left(k\right)=\left\{\begin{array}{cc}100& u_2\left(k\right)>100\\ u_2\left(k\right)+v_2\left(k\right)& -100\&le;u_2\left(k\right)\&le;100\\ -100& u_2\left(k\right)<-100\end{array}\right.$

The estimated value of source signal is y=W ₁G (u (w ₂X)).

If x _mBe m input source signal; If w _Jm ²Be the weights from the m input layer to the j hidden layer, 2 are from the input layer to the hidden layer; ${\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="A20081001141900162.png,A20081001141900171.png"\; state="\{\{state\}\}">w</figure-callout>}}_2=\left(w_{\mathrm{jm}}^2\right)$ Expression is by w _Jm ²The matrix of forming.

If u _jIt is the output valve of j hidden layer; w _Nj ¹Be the weights from j hidden layer to n output layer, 1 is from the hidden layer to the output layer; y _nBe n output valve. ${\mathrm{<figure-callout\; id="1"\; label="W"\; filenames="A20081001141900162.png,A20081001141900171.png"\; state="\{\{state\}\}">W</figure-callout>}}_1=w_{\mathrm{nj}}^1.$

If the separation matrix W on nxn rank, then it is output as

y(t)＝Wx(t)＝WAs(t)＝Gs(t)

According to blind source disjunctive model, error e=s-y of source signal s and its estimated signal y as noise signal, is set up the signal to noise ratio (S/N ratio) function of restoring signal s: $\mathrm{SNR}=10\log \frac{s*s^T}{e*e^T}=\frac{s*s^T}{(s-y)*{(s-y)}^T}.$

Because source signal is unknown, it is as follows therefore to obtain maximum SNR objective function with the running mean y replacement source signal of estimated signal: $F\left(y\right)=\mathrm{SNR}=10\log \frac{y*y^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}.$

Y=Wx in the formula, y=Wx, x are the signal of mixed signal x after running mean is handled.The maximum signal to noise ratio objective function can be write as following form:

$F(W,x)=10\log \frac{y{*y}^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}$

$=10\log \frac{\mathrm{Wx}*x^T{W}^T}{(s-y)*{(s-y)}^T}$

$=10\log \frac{{\mathrm{WCW}}^T}{W\stackrel{\&OverBar;}{C}{W}^T}$

$=10\log \frac{V}{U}$

The gradient of the function F of hence one can see that separation matrix W is $\frac{\&PartialD;F}{\&PartialD;W}=\frac{2W}{V}C-\frac{2W}{U}\stackrel{\&OverBar;}{C},$ Can try to achieve separation matrix

Separation matrix and source signal multiply each other and obtain final separation signal expression formula

The first machine is responsible for the collection of signal, and the second machine is responsible for the separating treatment of signal, presses the software flow working procedure in the second machine.Final output image such as Figure 10.Expression formula is:

y＝65+24.125*sin(x)+3.931*sin(3*x-3.32)+2.092*sin(5*x+2.11)+1.951*sin(7*x+0.12)+1.692*sin(9*x+0.26)+1.393*sin(11*x-5.61)+1.081*sin(13*x-2.45)+1.098*sin(15*x+0.15)+1.08*sin(17*x-5.24)+1.005*sin(19*x+4.18)。

Claims (4)

1, a kind of method that the power system voltage current signal is separated is characterized in that relying on device, may further comprise the steps:

Step 1, beginning;

Step 2, collection source signal

Use device is gathered voltage, the electric current of electric system;

Step 3, determine the separated network structure

After collecting signal, select network structure to separate according to the number of the aliasing signal that will separate;

Step 4, the non-linear aliasing model matrix of setting are isolated the estimated signal of signal as source signal

Exist n from the statistics of signal source signal independently, s _i(t), the mixed signal x of i=1～n and similar number _i(t), the mixing of signal is a linear instantaneous, and mixed signal is

$x_i\left(t\right)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_{\mathrm{ij}}{s}_j\left(t\right),$ A in the formula _IjBe mixing constant;

Estimated signal y (t)=Wx (t)=WAs (t)=Gs (t)

Overall situation matrix G=PD

Wherein: P is a nxn dimension permutation matrix, and D is a nxn dimension diagonal matrix,

Y (t)=PDs (t) then at first is that to set up one be the objective function F (W) of argument with W, if certain

F (W) is reached greatly or minimal value, should

Be required separating;

The error of step 5, calculating source signal and its estimated signal

Error e=s-y of source signal s and its estimated signal y as noise signal;

Step 6, set up the signal to noise ratio (S/N ratio) function of restoring signal

Set up the signal to noise ratio (S/N ratio) function of restoring signal s: $\mathrm{SNR}=10\log \frac{s*s^T}{e*e^T}=\frac{s*s^T}{(s-y)*{(s-y)}^T};$

Step 7, set up the maximum signal to noise ratio objective function

Running mean y with estimated signal replaces source signal, and it is as follows to obtain the maximum signal to noise ratio objective function:

$F\left(y\right)=\mathrm{SNR}=10\log \frac{y*y^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T},$

Y=Wx in the formula, y=Wx, x are the signal of mixed signal x after running mean is handled;

Step 8, optimization aim function

The following form of maximum signal to noise ratio objective function:

$F(W,x)=10\log \frac{y*y^T}{(\stackrel{\&OverBar;}{y}-y)*{(\stackrel{\&OverBar;}{y}-y)}^T}$

$=10\log \frac{\mathrm{Wx}*x^T{W}^T}{(s-y)*{(s-y)}^T}$

$=10\log \frac{\mathrm{WC}{W}^T}{W\stackrel{\&OverBar;}{C}{W}^T}$

$=10\log \frac{V}{U}$

Step 9, calculating separation matrix gradient

The gradient of the function F of separation matrix W is $\frac{\&PartialD;F}{\&PartialD;W}=\frac{2W}{V}C-\frac{2W}{U}\stackrel{\&OverBar;}{C},$ Try to achieve separation matrix

Step 10, separation matrix and source signal multiply each other and obtain final separation signal expression formula

Separation matrix and source signal multiply each other and obtain final separation signal expression formula;

Step 11, being delivered to host computer and showing with the first-harmonic of voltage and current signal and harmonic wave thereof.

2, according to the described a kind of method that the power system voltage current signal is separated of claim 1, it is characterized in that determining in the described step 3 the separated network structure, when there are 6 road signals in system, it is corresponding with the input source signal to construct 6 sub-PID neuroids, network structure promptly has 6 sub-PID neuroids, the structure that can set each sub-network is 2-2-2 or 3-3-3, single PI neural network is 3 layers of feedforward network, its structure is 2-2-2, the input layer of network is made of two neurons, the blind signal of aliasing of corresponding two inputs; Hidden layer also has two neurons, corresponds respectively to ratio (P), two parts of integration (I); The output layer of network is exactly the signal of two separation, the forward calculation of network realizes the calculation task of decomposition, the inverse algorithms of network is realized the self-adaptation adjustment of pid parameter, and when more signal input, it is corresponding with the input source signal to construct a plurality of sub-PID neuroids.

3, a kind of a kind of device that method adopted that the power system voltage current signal is separated as claimed in claim 1, it is characterized in that forming by slave computer and host computer, slave computer comprises sensor, the signal condition module, the A/D modular converter, the double mcu system unit, memory module, Keysheet module, communication module, sensor wherein, the signal condition module, the A/D modular converter, the double mcu system unit connects successively, memory module, the keyboard operation unit module, communication module links to each other with the double mcu system unit respectively, and host computer links to each other with communication module;

Described sensor comprises the voltage transformer (VT) summation current transformer, gather the voltage and current of electric system, be input to the signal condition module, the signal condition module is regulated its size by proper proportion and is made it to become the treatable amplitude range of device, the electric current and voltage of A/D modular converter acknowledge(ment) signal conditioning module conditioning is converted into the digital signal that the double mcu system unit can be discerned; The double mcu system unit comprises first machine and second machine, and the first machine is responsible for the control data gatherer process, second machine separation signal, and the first machine links to each other by conversion chip with the second machine; The first machine is connected with memory module by another conversion chip, and the second machine directly links to each other with memory module, memory module embedded software program, second machine software program for execution; Keysheet module is connected with the double mcu system unit by parallel interface, is used for the various correlation parameters of input system;

The identification result that described communication module is handled the double mcu system unit is sent to host computer, shows correlation parameter and Signal Separation image at last in the host computer liquid crystal display.

4, a kind of device that method adopted that the power system voltage current signal is separated as claimed in claim 3 is characterized in that the serial line interface of communication module adopts RS232 agreement and host computer Data transmission.

Images