CN101715257B - Intelligent controller of electric furnace electrode - Google Patents

Abstract

The invention relates to an intelligent controller of an electric furnace electrode. A plurality of digital/analog input and output channel modules are spliced with a single-board computer, and exterior terminals of the modules are connected with an electric furnace electrode unlocking operation switch, an elevating operation switch, a control valve of a locking device, a control valve of an elevating device, an upper and a lower space-limiting switches of a furnace cover, a manual/automatic switching switch, a current and voltage detection end, a current setting potentiometer, a transformer reactor shift and a fault comprehensive signal of an electric furnace control system. The electrode adjusting calculation method consists of an automatic arcing algorithm of equal arcing distance, a window current integrator overcurrent control algorithm, and a near reverse internal model control method based on a LSSVM-NARMA model. The electrode adjusting algorithm also comprises a PID control method, provides modeling data for the reverse internal model control method, is an alternative control method, and has the characteristic of quick and accurate electrode adjustment. The invention can significantly shorten the smelting period, save the electric energy, reduce the energy consumption, and is suitable for organizations of electric furnace smelting, and the like.

Description

Intelligent controller of electric furnace electrode

Technical field

What the present invention relates to is controller, particularly the contrary control method of a kind of intelligent internal mold of electric furnace steel making and integrated control device.

Background technology

In recent years, electric-arc furnace steelmaking has become one of main steel-making mode.The rise fall of electrodes regulating system is the important component part of electric arc furnaces, and its working effect quality directly affects output, quality and the energy resource consumption of steel.Therefore, seek a kind of simple and effective method so that the electrode regulating system can have preferably performance in smelting process, production has great importance for iron and steel.And the electrode regulating system is the time-varying system of a multivariate, non-linear, strong coupling, and random disturbance is very serious.The main thereof using PID control of actual industrial process, the PID control structure is simple, be easy to realize and have preferably robustness, but serious for coupling, the time electric arc furnaces that becomes to picture, stability and the control accuracy of PID control often can not be guaranteed, so that electrode work is difficult for stablizing, cause power consumption to increase decrease in efficiency, area power grid voltage fluctuation.

Non-linear autoregressive sliding model (NARMA) can be described the input/output relation of nonlinear system accurately.Even but know the NARMA model of system, because the nonlinear relationship of input and output is calculated control rate also very difficult, for this problem, recent years, existing many scholars have proposed several strategies, and use in several nonlinear Control problems.But mostly adopt neural network to be similar to non-linear partial, the training method based on gradient is mostly adopted in the training of neural network, has unavoidably the problem of local minimum point, and the structure of neural network also is difficult to determine.

In addition, electric furnace work relates to many control functions, controls, changes electrode control, reactor control, buggy ladle control, oxygen blast control or Argon control etc. such as Electrode control, bell control, charging control, cooling water control, transformer control, Hydraulic Station.At present, generally adopt whole stove PLC control design proposal both at home and abroad, limited the development of electric furnace control technology, some senior control methods are difficult to apply in a flexible way at PLC, especially as electric furnace control the core--Electrode control does not also have special product, in order to improve the control level of electric furnace, make its standardization and have universality, need the special electric furnace electrode lifting controller of design.

Summary of the invention

The objective of the invention is the problems referred to above of existing for the rise fall of electrodes regulating system of present electric-arc furnace steelmaking, and provide a kind of optimization, miniaturization, the embedded electric furnace intelligent electrode control device of nesting structural embedded control.

The technical scheme that adopts for the realization of above-mentioned purpose:

Intelligent controller of electric furnace electrode comprises hardware and software two parts, and hardware components comprises casing and is installed in the liquid crystal touch screen on the casing, the single board computer of the interior installing of casing, a plurality of digital-to-analog amount IO channel modules;

Described a plurality of digital-to-analog amount IO channel module is connected with the corresponding end pin of the cpu chip of single board computer by data bus, address bus and control bus; The associated end pin of cpu chip is connected with touch-screen display card, keyboard, mouse, USB interface and Ethernet card by wire respectively;

The end pin of above-mentioned a plurality of digital-to-analog amount IO channel modules respectively with the A of the electric furnace secondary side of furnace ' s control system, B, the test side of C phase current and voltage, with electric furnace A, B, the electrode of C phase rises, the step-down operation switch leads, with transformer, the gear terminals that reactor is corresponding, with bell upper limit position switch and lower position switch lead-in wire, with A, B, the pot output terminal that the C three-phase current is set, with A, B, C phase rise fall of electrodes unlocking operation switch leads, with the furnace ' s control system starting switch, with furnace ' s control system fault comprehensive signal (the fault comprehensive signal is provided by other opertaing devices in the furnace ' s control system, and the present invention does not relate to the particular content of furnace ' s control system), be connected with the lead end of manual/auto change-over switch etc.;

The end pin of above-mentioned a plurality of digital-to-analog amount IO channel modules respectively A, B, the C phase rise fall of electrodes locking device of corresponding and furnace ' s control system the operation valve end, be connected with A, B, C phase electrode lifting device (being generally hydraulic means) operation valve end.

Above-mentioned software section comprises setting parameter and display module, process simulation and display module, and process variable trend monitors module, alarm module and electrode control algolithm; Described setting parameter and display module, process simulation and display module, what process variable trend monitored module and alarm module employing is known technology;

Above-mentioned electrode control algolithm is another pith of the present invention, and the electrode control algolithm mainly comprises electrode regulating algorithm, automatic arc striking control algolithm and excess current control algorithm.The electrode regulating algorithm comprises two kinds of control methods, PID control method and contrary internal model control method, and the PID control method also is control method for subsequent use for contrary internal model control method provides modeling data, contrary internal model control method is main method.The overall control block diagram as shown in Figure 2.When electric furnace was controlled automatically, electrode control algolithm shown in Figure 2 was effective.The rising or falling speed of electrode will be regulated according to following situation:

During initial use electric furnace, soft switch k1 is in a position shown in Fig. 2, adopt conventional PID control method, (electric furnace is the metallurgical equipment of batch work after smelting some batches, comprise charging, smelt, tap several steps), collect all data and set up approximate inverse internal mold model, obtain the approximate inverse Internal Model Control Algorithm.In the control of each batch, the work of electrode control algolithm is as follows: at first, soft switch k1 is in a position among Fig. 2, and soft switch k2 is in the d position, and soft switch k3 is in the e position, and soft switch k4 is in the h position, carries out the control of automatic arc striking algorithm.After electrode dropped to starting the arc position, soft switch k3 was in the f position, and other soft position of the switch are constant, carried out pid algorithm and automatically controlled.No matter be in automatic arc striking algorithm control stage or pid algorithm control stage, when output window current integration value exceeds the upper limit, soft switch k3 will be in e position (otherwise still being in the f position), start the excess current control algorithm; During operative employee's manual operation electrode (according to on-the-spot needs), soft switch k4 will be in g position (inoperation still is in the h position), and manually controlled quentity controlled variable is preferential.

After obtaining the approximate inverse Internal Model Control Algorithm, soft switch k1 is in the b position shown in Fig. 2, adopts approximate inverse internal mode controller method.In the control of each batch, the work of electrode control algolithm is as follows: at first, soft switch k1 is in the b position among Fig. 2, and soft switch k2 is in the d position, and soft switch k3 is in the e position, and soft switch k4 is in the h position, carries out the control of automatic arc striking algorithm.After electrode dropped to starting the arc position, soft switch k3 was in the f position, and other soft position of the switch are constant, carried out pid algorithm and automatically controlled.No matter be at automatic arc striking algorithm control stage or approximate inverse Internal Model Control Algorithm, when output window current integration value exceeds the upper limit, soft switch k3 will be in e position (otherwise still being in the f position), start the excess current control algorithm; During operative employee's manual operation electrode (according to on-the-spot needs), soft switch k4 will be in g position (inoperation still is in the h position), and manually controlled quentity controlled variable is preferential.

Above-mentioned electric furnace approximate inverse Internal Model Control Algorithm is set up and is comprised following three steps:

The first step is for approximate least square method supporting vector machine (LSSVM) model of the NARMA that obtains electric furnace system, the NARMA mechanism model of first analytic system;

The hydraulic means of control electrode lifting and electric furnace main circuit are partly regarded as broad sense, and controlled (three-phase electrode of electric furnace links to each other with Circuit Fault on Secondary Transformer as the conductor upper end to picture, the lower end keeps certain distance to consist of the electric furnace major loop with steel scrap to be refined or molten steel), it is controlled to the picture model that the below sets up broad sense.

Hydraulic means can be similar to regards a third order PLL joint as, and the transport function form is:

Then: $l_j=\frac{b_0}{a_0{s}^3+a_1{s}^2+a_{2}s}{u}_j,j=a,b,c;---\left(1\right)$

That is: $a_0{l}_j^{\left(3\right)}+a_1{l}_j^{\left(2\right)}+a_2{l}_j^{\left(1\right)}=b_0{u}_j,j=a,b,c;---\left(2\right)$

u _jBe each phase control signal, l _jBe each phase arc length.

Arc Modelling adopts the kohle model, and its form is as being: Z _Arc=R _Arc+ j*X _ArcWherein $X_{\mathrm{arc}}=a*R_{\mathrm{arc}}+b*R_{\mathrm{arc}}^2,$ R _Arc=l*R _Per.Z _ArcBe arc impedance, R _PerArc resistance value for per unit length.L is arc length, and a gets different values with b, to reflect the different smelting stages.

Furnace transformer is selected the Yd11 connection, utilizes the transformer correlation theory, and in conjunction with electric arc furnaces main circuit electrical system, the effective value that can release the secondary lateral of electric stove transformer electric current is:

$\left\{\begin{array}{c}{i}_a=\left|\right|\frac{{\stackrel{\·}{U}}_A(Z_k/3n-n*Z_b)-{\stackrel{\·}{U}}_B(Z_k/3n-n*Z_c)}{Z_{\mathrm{Kabc}}}\left|\right|\\ i_b=\left|\right|\frac{{\stackrel{\·}{U}}_B(Z_k/3n-n*Z_c)-{\stackrel{\·}{U}}_C(Z_k/3n-n*Z_a)}{Z_{\mathrm{Kabc}}}\left|\right|\\ i_c=\left|\right|\frac{{\stackrel{\·}{U}}_C(Z_k/3n-n*Z_a)-{\stackrel{\·}{U}}_A(Z_k/3n-n*Z_b)}{Z_{\mathrm{Kabc}}}\left|\right|\end{array}\right.---\left(3\right)$

Wherein:

Z _Kabc＝(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _b)+(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _c)+(Z _K/3n-n*Z _b)(Z _K/3n-n*Z _c)

Z _j＝R _d+j*X _d+l _j*R _per+j*(a*l _j*R _per+b*(l _j*R _per) ²)，j＝a，b，c；

Wherein: R _dAnd X _dBe short net impedance and induction reactance.N is the furnace transformer no-load voltage ratio.

Be furnace transformer primary side phase voltage.Z _K: the transformer leakage impedance.

Adopt n rank approximation method: ${\mathrm{Ti}}_j^{\left(1\right)}=i_j(k+1)-i_j\left(k\right),$ $T^2{i}_j^{\left(2\right)}=i_j(k+1)-{2i}_j\left(k\right)+i_j(k-1),$

$T^3{i}_j^{\left(3\right)}=i_j(k+1)-{3i}_j\left(k\right)+{3i}_j(k-1)-i_j(k-2),$ J=a, b, c; T is the sampling period.Convolution (1)-(3),

Obtain the broad sense electric arc furnaces to as the NARMA model being:

i(k+1)＝f[ω(k)，u(k)]+v(k)(4)

Wherein:

f[ω(k)，u(k)]＝[f _a[ω(k)，u(k)]，f _b[ω(k)，u(k)]，f _c[ω(k)，u(k)]] ^T

i(k+1)＝[i _a(k+1)，i _b(k+1)，i _c(k+1)] ^T u(k)＝[u _a(k)，u _b(k)，u _c(k)] ^T

ω(k)＝[i _a(k)，i _a(k-1)，i _a(k-2)，i _b(k)，i _b(k-1)，i _b(k-2)，i _c(k)，i _c(k-1)，i _c(k-2)] ^T

v(k)＝[v _a(k)，v _b(k)，v _c(k)] ^T

F is the nonlinear function vector, f _a, f _b, f _cBe nonlinear function.V (k) is vectorial for disturbing, and establishes || v (k) || and≤v ₀.

Second step adopts LSSVM that broad sense is carried out identification to picture

The kernel function type of LSSVM adopts gaussian kernel function, and functional form is: $K(x_i,x_j)=\exp \left(\frac{-\left|\right|x_i-x_j\left|\right|}{{2\mathrm{\σ}}^2}\right).$ Three-phase current adopts three LSSVM models to carry out identification.Given training dataset, { [ω (k), u (k)], i (k+1) } input data are: [ω (k), u (k)], and the output data are: i (k+1), adopt the learning algorithm of LSSVM, and obtain LSSVM NARMA model and be:

${\hat{i}}_j(k+1)=f[\mathrm{\ω}\left(k\right),u\left(k\right)]$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\Σ}}}{\mathrm{\α}}_{t}K\left[\right[\mathrm{\ω}\left(k\right),u\left(k\right)],[\mathrm{\ω}\left(t\right),u\left(t\right)\left]\right]+b$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\Σ}}}{\mathrm{\α}}_t\exp \left(\frac{-\left|\right|[\mathrm{\ω}\left(k\right),u\left(k\right)]-[\mathrm{\ω}\left(t\right),u\left(t\right)]\left|\right|}{{2\mathrm{\σ}}^2}\right)+b---\left(5\right)$

j＝a，b，c；

SV expresses support for the number of vector machine, α _iWith b be model parameter.

For NARMA model (4), constantly carry out Taylor at k-1 and launch to obtain approximate model:

i(k+1)＝i(k)+f ₁*Δu(k)+R(k)+v(k)(6)

Wherein: ${\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="H2009102200497E00011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1=\frac{\&PartialD;f[\mathrm{\&omega;}\left(k\right),u\left(k\right)]}{\&PartialD;u\left(k\right)}{|}_{u\left(k\right)=u(k-1),\mathrm{\&omega;}\left(k\right)=\mathrm{\&omega;}(k-1)}$

$f_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]=\frac{{\&PartialD;}^{2}f[\mathrm{\&omega;}(k-1),u\left(k\right)]}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

$R\left(k\right)=\frac{{\left[\mathrm{\&Delta;u}\left(k\right)\right]}^T{f}_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]\mathrm{\&Delta;u}\left(k\right)}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

ζ=[ζ _a, ζ _b, ζ _c] ^T, and u _j(k-1)≤ζ _j≤ u _j(k), j=a, b, c;

By formula (6), ignore R (k) and v (k), the Taylor approximate model that obtains system is:

i(k+1)＝i(k)+f ₁*Δu(k)(7)

The 3rd step, the approximate inverse control algolithm

By the output of LSSVM NARMA model (5) derivative being asked in corresponding input obtains

${\hat{f}}_1=\left[\begin{array}{ccc}-\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)\\ \text{-}\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)\\ -\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)\end{array}\right]---\left(8\right)$

Wherein: K _j(k-1)=K _j[Q (k-1), Q (t)], j=a, b, c;

And then obtain the computing formula of approximate inverse control algolithm according to formula (7):

u _j(k)＝u _j(k-1)+Δu _j(k) j＝a，b，c； (9)

${\mathrm{\&Delta;u}}_j\left(k\right)={{\hat{f}}_1}_j^T{({\hat{f}}_1{{\hat{f}}_1}^T+\mathrm{\&alpha;})}^{-1}*(r(k+1)-i\left(k\right))$ When | Δ u _j(k) |≤δ _j

Δ u _j(k)=δ _jSign[Δ u _j(k)] as | Δ u _j(k) |＞δ _j

Wherein: r (k+1) is current setting value, r (k+1)=[r _a(k+1), r _b(k+1), r _c(k+1)] ^T. δ=[δ _a, δ _b, δ _c,] ^T,

δ _jBe arithmetic number. ${\hat{f}}_1={[{\hat{f}}_{1a},{\hat{f}}_{1b},{\hat{f}}_{1c}]}^T.\mathrm{\&alpha;}=\mathrm{diag}({\mathrm{\&alpha;}}_a,{\mathrm{\&alpha;}}_b,{\mathrm{\&alpha;}}_c),$ α _jBe very little arithmetic number.Sign is sign function.

By prior process as can be known, uncertainty is that inevitably large uncertainty will reduce the performance of control system in real system.Weaken uncertainty with the internal mold strategy, when uncertainty exists, according to model and actual output error design control and compensation amount Δ u _c(k), adopt simultaneously robust filter F (z) to weaken uncertainty.

${\mathrm{\&Delta;u}}_{\mathrm{cj}}\left(k\right)=-{{{\hat{f}}_1}_j^T({\hat{f}}_1{{\hat{f}}_1}^T+\mathrm{\&alpha;})}^{-1}*F\left(z\right)*(R\left(k\right)+v\left(k\right))$ When | Δ u _Cj(k) |≤δ _Cj(10)

Δ u _Cj(k)=δ _CjSign[Δ u _Cj(k)], as | Δ u _Cj(k) |＞δ _Cj

J=a in the formula, b, c; δ _c=[δ _Ca, δ _Cb, δ _Cc], δ _CjBe arithmetic number, sign is sign function, and R (k)+v (k) uses

Replace.

To sum up must be with the approximate inverse Internal Model Control Algorithm of compensation:

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) (11)

Described excess current control algorithm is: the material that prevents from collapsing waits and causes instantaneous short circuit electrode mistuning joint to occur, whenever, design mutually a window integrator, when the current integration value surpasses the maximum limit definite value, the work of excess current control algorithm, so that excess current is worth larger, electrode speed raises faster.Like this, by promoting fast single rather than a plurality of relevant electrodes, correct overcurrent condition, the vibration that brings to avoid the electrode coupling impact is regulated, and avoids the overload of furnace transformer and electrode.Specific algorithm is:

$S_{\mathrm{nj}}=\underset{k-n-1}{\overset{k}{\mathrm{\&Sigma;}}}{i}_j\left(k\right)$

u _j(k)＝K _tS _nj S _nj≥L _max (12)

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) S _nj＜L _max

In the formula, n is window integrator length, K _tBe overcurrent controller scale-up factor, L _MaxBe current integration maximum limit definite value.

Described automatic arc striking algorithm is: when automatically moving, at first make three-phase electrode automatically drop to starting the arc position, then automatically adjust.Control method is: during automatic arc striking, A phase electrode descends automatically, and rising TA stops second after the A phase voltage drops to predetermined voltage, then B phase electrode descends automatically, rising TB stops second after the B phase voltage drops to predetermined voltage, and last C phase electrode descends automatically, and rising TC stops second after the C phase voltage drops to predetermined voltage, TA, TB, TC are obtained by rate pattern, Ti=L/Vi, in the formula, i=A, B, C; L is for preparing striking-distance; Vi is the electrode ascending velocity.

Principle of work of the present invention:

After the energising of this intelligent controller of electric furnace electrode, this intelligent controller of electric furnace electrode just enters wait state, and when the furnace ' s control system starting switch was effective, control system was started working.If play for the first time machine, need by prompting by touch screen operation interface initialization furnace parameter, comprise tonnage, electrode diameter, short network parameters, electrode speed, the corresponding rated voltage electric current of transformer gear, the corresponding reactance of reactor gear, pid control parameter, contrary internal model control parameter (furnace parameter also can at any time by the operation interface setting).

Rise fall of electrodes operation minute manual and automatic dual mode is when the input signal from hand/automatic key is low level, manually effectively.At this moment, the both separately liftings of A, B, C three-phase electrode, also simultaneously lifting is by the rise fall of electrodes signal deciding.When the rise fall of electrodes signal is high level, rise fall of electrodes release control signal is in the high level situation, and electrode is namely with maximal rate (can according to the customer requirements adjustment) lifting, when the rise fall of electrodes signal is low level, and the electrode stop motion.

When the input signal from hand/automatic switchover key is high level, automatically effectively.Automatically during operation, when system failure integrated signal without the time, rise fall of electrodes release control signal is automatically effective, automatic arc striking at first, A phase electrode descends automatically, and rising TA stops second after the A phase voltage drops to predetermined voltage, then B phase electrode descends automatically, and rising TB stops second after the B phase voltage drops to predetermined voltage, and last C phase electrode descends automatically, rising TC stops second after the C phase voltage drops to predetermined voltage, and TA, TB, TC are obtained by rate pattern, Ti=L/Vi, in the formula, i=A, B, C; L is for preparing striking-distance; Vi is the electrode ascending velocity.If then pid control algorithm (for collecting modeling data) is then selected in initial use according to electric furnace, according to the flame current of transformer gear, flame current setting value and actual measurement, the position that arc voltage is regulated electrode automatically; If modeling data is collected and is finished, obtain formula (5) model and approximate inverse Internal Model Control Algorithm formula (11), then select approximate inverse Internal Model Control Algorithm formula (11) control.No matter be in automatic arc striking algorithm control stage or pid algorithm or approximate inverse Internal Model Control Algorithm stage, when window current integration value exceeds the upper limit, start the excess current control algorithm, control by formula (12); During operative employee's manual operation electrode (according to on-the-spot needs), manually controlled quentity controlled variable is preferential.Automatically during operation, when system failure integrated signal sometimes, auto-programming is not worked, manual process work.

Characteristics of the present invention:

This intelligent controller of electric furnace electrode is by casing and be installed in the single board computer of installing in liquid crystal touch screen on the casing, the casing, a plurality of digital-to-analogs inputs, output channel module form, and is a kind of optimization, miniaturization, nesting structural embedded control.

This intelligent controller of electric furnace electrode has designed flexibly electrode control correlation parameter and has set and function for monitoring, can realize by touch-screen (this locality), also can realize (long-range, by client development) by the engineer station by communication.Local function specifically comprises setting parameter and Presentation Function, process simulation Presentation Function, process variable trend function for monitoring, warning function.

The electrode control algolithm is important content of the present invention, and the automatic arc striking algorithm is determined the striking-distance of three-phase electrode by the voltage drop method, so that the electrode starting the arc is steady; Window integrator excess current control algorithm is corrected overcurrent condition by the single rather than a plurality of relevant electrodes of fast lifting, avoids the electrode coupling impact and the vibration that brings is regulated; The approximate inverse Internal Model Control Algorithm only needs an approximate LSSVM model, avoided identification inversion model computational complexity, control rate is calculated simple, is easy to actual realization, electrode regulating has fast and accurately characteristic, is the key that shortens smelting cycle, saves energy, reduction consumption of electrode.

Design science of the present invention, reasonable, original, compact overall structure, volume are little, low-power consumption, with low cost, stable performance, reliable, environmental suitability and practicality are stronger, have preferably development prospect.The relevant devices such as suitable relevant electric furnace smelting use.

Description of drawings

Fig. 1 is contour structures schematic diagram of the present invention.

Fig. 2 is hardware block diagram of the present invention.

Fig. 3 is control loop block diagram of the present invention.

Embodiment

Intelligent controller of electric furnace electrode comprises hardware and software two parts, and hardware components comprises casing 1 and is installed in single board computer, a plurality of digital-to-analog amount IO channel module that is equipped with in liquid crystal touch screen 2 on the casing 1, the casing 1.

Single board computer adopts the PC/104-1621CLDN model, digital-to-analog amount IO channel module adopts with the input of 16 railway digitals, the output of 16 railway digitals and 6 tunnel analog output channel module 104-A726, carry out stacking-type with 32 road analog input channel 104-32ADT and peg graft, liquid crystal touch screen 2 adopts TPC-064.The corresponding end pin of the cpu chip of single board computer connects hard disk, DVD drive, keyboard, mouse, USB interface, touch-screen display card and Ethernet card by wire respectively, and total interface is installed on the casing outside.

Described a plurality of digital-to-analog amount IO channel module comprises that an I/O module model is that 104-32ADT, the 2nd I/O module model are that 104-A726, the 3rd I/O module are for subsequent use, and above-mentioned first, second, and third I/O module is connected with the corresponding end pin of the cpu chip of single board computer PC/104 by data bus, address bus and control bus respectively;

An above-mentioned I/O module end pin respectively A, B, C phase current and the voltage of corresponding and electric furnace secondary side the test side, be connected with the pot output terminal of A, B, the setting of C three-phase current, gear terminals corresponding with transformer, reactor.

Above-mentioned the 2nd I/O module end pin respectively corresponding and electric furnace A, B, C phase electrode rising, step-down operation switch leads, with the lead end of manual/auto change-over switch, be connected with A, B, C phase lifting unlocking operation switch; With the operation valve end of A, B, C phase rise fall of electrodes locking device, with A, B, C phase electrode lifting device (being generally hydraulic means) operation valve end, with bell upper and lower limit bit switch lead-in wire, with the furnace ' s control system starting switch, be connected with system failure integrated signal (other opertaing devices provide);

Above-mentioned the 3rd I/O module is for subsequent use; The A of above-mentioned electric furnace secondary side, B, the test side of C phase current and voltage, A, B, the pot output terminal that the C three-phase current is set, the gear terminals of transformer and reactor, on the bell, the lower position switch lead-in wire, with the furnace ' s control system starting switch, the system failure, electric furnace A, B, the electrode of C phase rises, the step-down operation switch, manual/auto change-over switch, A, B, C phase lifting unlocking operation switch, A, B, operation valve and the A of C phase rise fall of electrodes locking device, B, C phase electrode lifting device (being generally hydraulic means) operation valve is all check point and the reference mark of furnace ' s control system.

Finish on the basis in the hardware connection, WINDOWS operating system and electric furnace electrode Based Intelligent Control application software are installed.

Above-mentioned software section adopts the C# language exploitation, comprises setting parameter and display module, process simulation and display module, and process variable trend monitors module, alarm module and electrode control algolithm; Setting parameter and display module, process simulation and display module, what process variable trend monitored module and alarm module employing is known technology;

Above-mentioned electrode control algolithm mainly comprises electrode regulating algorithm, automatic arc striking control algolithm and excess current control algorithm.The electrode regulating algorithm comprises two kinds of control methods, PID control method and contrary internal model control method, and the PID control method also is control method for subsequent use for contrary internal model control method provides modeling data, contrary internal model control method is main method.

Above-mentioned electric furnace approximate inverse Internal Model Control Algorithm is set up and is comprised following three steps:

The first step is for approximate least square method supporting vector machine (LSSVM) model of the NARMA that obtains electric furnace system, the NARMA mechanism model of first analytic system;

The hydraulic means of control electrode lifting and electric furnace main circuit are partly regarded as broad sense, and controlled (three-phase electrode of electric furnace links to each other with Circuit Fault on Secondary Transformer as the conductor upper end to picture, the lower end keeps certain distance to consist of the electric furnace major loop with steel scrap to be refined or molten steel), it is controlled to the picture model that the below sets up broad sense.

Hydraulic means can be similar to regards a third order PLL joint as, and the transport function form is:

Then: $l_j=\frac{b_0}{a_0{s}^3+a_1{s}^2+a_{2}s}{u}_j,j=a,b,c;---\left(1\right)$

That is: $a_0{l}_j^{\left(3\right)}+a_1{l}_j^{\left(2\right)}+a_2{l}_j^{\left(1\right)}=b_0{u}_j,j=a,b,c;---\left(2\right)$

u _jBe each phase control signal, l _jBe each phase arc length.

Arc Modelling adopts the kohle model, and its form is as being: Z _Arc=R _Arc+ j*X _ArcWherein $X_{\mathrm{arc}}=a*R_{\mathrm{arc}}+b*R_{\mathrm{arc}}^2,$ R _Arc=l*R _Per.Z _ArcBe arc impedance, R _PerArc resistance value for per unit length.L is arc length, and a gets different values with b, to reflect the different smelting stages.

Furnace transformer is selected the Yd11 connection, utilizes the transformer correlation theory, and in conjunction with electric arc furnaces main circuit electrical system, the effective value that can release the secondary lateral of electric stove transformer electric current is:

$\left\{\begin{array}{c}{i}_a=\left|\right|\frac{{\stackrel{\&CenterDot;}{U}}_A(Z_k/3n-n*Z_b)-{\stackrel{\&CenterDot;}{U}}_B(Z_k/3n-n*Z_c)}{Z_{\mathrm{Kabc}}}\left|\right|\\ i_b=\left|\right|\frac{{\stackrel{\&CenterDot;}{U}}_B(Z_k/3n-n*Z_c)-{\stackrel{\&CenterDot;}{U}}_C(Z_k/3n-n*Z_a)}{Z_{\mathrm{Kabc}}}\left|\right|\\ i_c=\left|\right|\frac{{\stackrel{\&CenterDot;}{U}}_C(Z_k/3n-n*Z_a)-{\stackrel{\&CenterDot;}{U}}_A(Z_k/3n-n*Z_b)}{Z_{\mathrm{Kabc}}}\left|\right|\end{array}\right.---\left(3\right)$

Wherein:

Z _Kabc＝(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _b)+(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _c)+(Z _K/3n-n*Z _b)(Z _K/3n-n*Z _c)

Z _j＝R _d+j*X _d+l _j*R _per+j*(a*l _j*R _per+b*(l _j*R _per) ²)，j＝a，b，c；

Wherein: R _dAnd X _dBe short net impedance and induction reactance.N is the furnace transformer no-load voltage ratio. Be furnace transformer primary side phase voltage.Z _K: the transformer leakage impedance.

Adopt n rank approximation method: ${\mathrm{Ti}}_j^{\left(1\right)}=i_j(k+1)-i_j\left(k\right),$ $T^2{i}_j^{\left(2\right)}=i_j(k+1)-{2i}_j\left(k\right)+i_j(k-1),$

$T^3{i}_j^{\left(3\right)}=i_j(k+1)-{3i}_j\left(k\right)+{3i}_j(k-1)-i_j(k-2),$ J=a, b, c; T is the sampling period.Convolution (1)-(3) obtain the broad sense electric arc furnaces to as the NARMA model being:

i(k+1)＝f[ω(k)，u(k)]+v(k) (4)

Wherein:

f[ω(k)，u(k)]＝[f _a[ω(k)，u(k)]，f _b[ω(k)，u(k)]，f _c[ω(k)，u(k)]] ^T

i(k+1)＝[i _a(k+1)，i _b(k+1)，i _c(k+1)] ^T u(k)＝[u _a(k)，u _b(k)，u _c(k)] ^T

ω(k)＝[i _a(k)，i _a(k-1)，i _a(k-2)，i _b(k)，i _b(k-1)，i _b(k-2)，i _c(k)，i _c(k-1)，i _c(k-2)] ^T

v(k)＝[v _a(k)，v _b(k)，v _c(k)] ^T

F is the nonlinear function vector, f _a, f _b, f _cBe nonlinear function.V (k) is vectorial for disturbing, and establishes || v (k) || and≤v ₀.

Second step is adopted LSSVM broad sense is carried out identification to picture

The kernel function type of LSSVM adopts gaussian kernel function, and functional form is: $K(x_i,x_j)=\exp \left(\frac{-\left|\right|x_i-x_j\left|\right|}{{2\mathrm{\&sigma;}}^2}\right).$ Three-phase current adopts three LSSVM models to carry out identification.Given training dataset, { [ω (k), u (k)], i (k+1) } input data are: [ω (k), u (k)], and the output data are: i (k+1), adopt the learning algorithm of LSSVM, and obtain LSSVM NARMA model and be:

${\hat{i}}_j(k+1)=f[\mathrm{\&omega;}\left(k\right),u\left(k\right)]$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t}K\left[\right[\mathrm{\&omega;}\left(k\right),u\left(k\right)],[\mathrm{\&omega;}\left(t\right),u\left(t\right)\left]\right]+b$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t\exp \left(\frac{-\left|\right|[\mathrm{\&omega;}\left(k\right),u\left(k\right)]-[\mathrm{\&omega;}\left(t\right),u\left(t\right)]\left|\right|}{{2\mathrm{\&sigma;}}^2}\right)+b---\left(5\right)$

j＝a，b，c；

SV expresses support for the number of vector machine, α _tWith b be model parameter.

For NARMA model (4), constantly carry out Taylor at k-1 and launch to obtain approximate model:

i(k+1)＝i(k)+f ₁*Δu(k)+R(k)+v(k) (6)

Wherein: ${\mathrm{<figure-callout\; id="1"\; label="f"\; filenames="H2009102200497E00011.png"\; state="\{\{state\}\}">f</figure-callout>}}_1=\frac{\&PartialD;f[\mathrm{\&omega;}\left(k\right),u\left(k\right)]}{\&PartialD;u\left(k\right)}{|}_{u\left(k\right)=u(k-1),\mathrm{\&omega;}\left(k\right)=\mathrm{\&omega;}(k-1)}$

$f_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]=\frac{{\&PartialD;}^{2}f[\mathrm{\&omega;}(k-1),u\left(k\right)]}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

$R\left(k\right)=\frac{{\left[\mathrm{\&Delta;u}\left(k\right)\right]}^T{f}_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]\mathrm{\&Delta;u}\left(k\right)}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

ζ=[ζ _a, ζ _b, ζ _c] ^T, and u _j(k-1)≤ζ _j≤ u _j(k), j=a, b, c;

By formula (6), ignore R (k) and v (k), the Taylor approximate model that obtains system is:

i(k+1)＝i(k)+f ₁*Δu(k) (7)

The 3rd step, the approximate inverse control algolithm

By the output of LSSVM NARMA model (5) derivative being asked in corresponding input obtains

${\hat{f}}_1=\left[\begin{array}{ccc}-\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)\\ \text{-}\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)\\ -\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)\end{array}\right]---\left(8\right)$

Wherein: K _j(k-1)=K _j[Q (k-1), Q (t)], j=a, b, c;

And then obtain the computing formula of approximate inverse control algolithm according to formula (7):

u _j(k)＝u _j(k-1)+Δu _j(k) j＝a，b，c； (9)

${\mathrm{\&Delta;u}}_j\left(k\right)={{{\hat{f}}_1}_j^T({\hat{f}}_1{{\hat{f}}_1}^T+a)}^{-1}*(r(k+1)-i\left(k\right))$ When | Δ u _j(k) |≤δ _j

Δ u _j(k)=δ _jSign[Δ u _j(k)] as | Δ u _j(k) |＞δ _j

Wherein: r (k+1) is current setting value, r (k+1)=[r _a(k+1), r _b(k+1), r _c(k+1)] ^T. δ=[δ _a, δ _b, δ _c,] ^T,

δ _jBe arithmetic number. ${\hat{f}}_1={[{\hat{f}}_{1a},{\hat{f}}_{1b},{\hat{f}}_{1c}]}^T\mathrm{\&alpha;}=\mathrm{diag}({\mathrm{\&alpha;}}_a,{\mathrm{\&alpha;}}_b,{\mathrm{\&alpha;}}_c),$ α _jBe very little arithmetic number.Sign is sign function.

By prior process as can be known, uncertainty is that inevitably large uncertainty will reduce the performance of control system in real system.Weaken uncertainty with the internal mold strategy, when uncertainty exists, according to model and actual output error design control and compensation amount Δ u _c(k), adopt simultaneously robust filter F (z) to weaken uncertainty.

${\mathrm{\&Delta;u}}_{\mathrm{cj}}\left(k\right)=-{{\hat{f}}_1}_j^T{({\hat{f}}_1{{\hat{f}}_1}^T+\mathrm{\&alpha;})}^{-1}*F\left(z\right)*(R\left(k\right)+v\left(k\right))$ When | Δ u _Cj(k) |≤δ _Cj(10)

Δ u _Cj(k)=δ _CjSign[Δ u _Cj(k)], as | Δ u _Cj(k) |＞δ _Cj

J=a in the formula, b, c; δ _c=[δ _Ca, δ _Cb, δ _Cc], δ _CjBe arithmetic number, sign is sign function, and R (k)+v (k) uses

Replace.

To sum up must be with the approximate inverse Internal Model Control Algorithm of compensation:

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) (11)

Described excess current control algorithm is: the material that prevents from collapsing waits and causes instantaneous short circuit electrode mistuning joint to occur, whenever, design mutually a window integrator, when the current integration value surpasses the maximum limit definite value, the work of excess current control algorithm, so that excess current is worth larger, electrode speed raises faster.Like this, by promoting fast single rather than a plurality of relevant electrodes, correct overcurrent condition, the vibration that brings to avoid the electrode coupling impact is regulated, and avoids the overload of furnace transformer and electrode.Specific algorithm is:

$S_{\mathrm{nj}}=\underset{k-n-1}{\overset{k}{\mathrm{\&Sigma;}}}{i}_j\left(k\right)$

u _j(k)＝K _tS _nj S _nj≥L _max (12)

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) S _nj＜L _max

In the formula, n is window integrator length, K _tBe overcurrent controller scale-up factor, L _MaxBe current integration maximum limit definite value.

Described automatic arc striking algorithm is: when automatically moving, at first make three-phase electrode automatically drop to starting the arc position, then automatically adjust.Control method is: during automatic arc striking, A phase electrode descends automatically, and rising TA stops second after the A phase voltage drops to predetermined voltage, then B phase electrode descends automatically, rising TB stops second after the B phase voltage drops to predetermined voltage, and last C phase electrode descends automatically, and rising TC stops second after the C phase voltage drops to predetermined voltage, TA, TB, TC are obtained by rate pattern, Ti=L/Vi, in the formula, i=A, B, C; L is for preparing striking-distance; Vi is the electrode ascending velocity.

After the energising of this intelligent controller of electric furnace electrode, this intelligent controller of electric furnace electrode just enters wait state, and when the furnace ' s control system starting switch was effective, control system was started working.If play for the first time machine, need to pass through touch screen operation interface initialization furnace parameter by prompting, comprise tonnage, electrode diameter, short network parameters, electrode speed, the corresponding rated voltage electric current of transformer gear, the corresponding reactance of reactor gear, pid control parameter, contrary internal model control parameter.

Rise fall of electrodes operation minute manual and automatic dual mode is when the input signal from hand/automatic key is low level, manually effectively.At this moment, the both separately liftings of A, B, C three-phase electrode, also simultaneously lifting is by the rise fall of electrodes signal deciding.When the rise fall of electrodes signal is high level, rise fall of electrodes release control signal is in the high level situation, and electrode is namely with maximal rate (can according to the customer requirements adjustment) lifting, when the rise fall of electrodes signal is low level, and the electrode stop motion.

When the input signal from hand/automatic switchover key is high level, automatically effectively.Automatically during operation, when system failure integrated signal without the time, rise fall of electrodes release control signal is automatically effective, according to the starting the arc of automatic arc striking algorithm.If then pid control algorithm (for collecting modeling data) is then selected in initial use according to electric furnace, according to the flame current of transformer gear, flame current setting value and actual measurement, the position that arc voltage is regulated electrode automatically; Collect all modeling datas after moving 10 batches, obtain formula (5) model and approximate inverse Internal Model Control Algorithm formula (11), then select approximate inverse Internal Model Control Algorithm formula (11) control.No matter be in automatic arc striking algorithm control stage or pid algorithm or approximate inverse Internal Model Control Algorithm stage, when output window current integration value exceeds the upper limit, start the excess current control algorithm, control by formula (12); During operative employee's manual operation electrode (according to on-the-spot needs), manually controlled quentity controlled variable is preferential.Automatically during operation, when system failure integrated signal sometimes, auto-programming is not worked, manual process work.

Claims (1)

1. intelligent controller of electric furnace electrode, comprise hardware and software two parts, hardware components comprise casing (1) and be installed in liquid crystal touch screen on the casing (1) and keyboard and mouse (2), casing (1) in single board computer and a plurality of digital-to-analog amount IO channel module of installing; Described a plurality of digital-to-analog amount IO channel module is connected with the corresponding end pin of the cpu chip of single board computer by data bus, address bus and control bus; The associated end pin of cpu chip is connected with liquid crystal touch screen, keyboard, mouse by wire respectively; It is characterized in that:

The end pin of above-mentioned a plurality of digital-to-analog amount IO channel modules respectively with the A of the electric furnace secondary side of furnace ' s control system, B, the test side of C phase current and voltage, with electric furnace A, B, the electrode of C phase rises, the step-down operation switch leads, with transformer, the gear terminals that reactor is corresponding, with bell upper limit position switch and lower position switch lead-in wire, A with the electric furnace secondary side, B, the pot output terminal that the C three-phase current is set, with A, B, C phase rise fall of electrodes unlocking operation switch leads, with the furnace ' s control system starting switch, with system failure integrated signal, be connected with the lead end of manual/auto change-over switch;

The end pin of above-mentioned a plurality of digital-to-analog amount IO channel modules respectively A, B, the C phase rise fall of electrodes locking device of corresponding and furnace ' s control system the operation valve end, be connected with A, B, C phase electrode lifting device operation valve end;

Above-mentioned software section comprises the electrode control algolithm; Above-mentioned electrode control algolithm mainly comprises electrode regulating algorithm, automatic arc striking control algolithm and excess current control algorithm; The electrode regulating algorithm comprises two kinds of control methods, PID control method and approximate inverse Internal Model Control Algorithm, the PID control method is the control method for subsequent use of approximate inverse Internal Model Control Algorithm for the approximate inverse Internal Model Control Algorithm provides modeling data, and the approximate inverse Internal Model Control Algorithm is main method;

Above-mentioned electric furnace approximate inverse Internal Model Control Algorithm is set up and is comprised following three steps:

The first step is for approximate least square method supporting vector machine (LSSVM) model of the NARMA that obtains electric furnace system, the NARMA mechanism model of first analytic system;

The hydraulic means of control electrode lifting and electric furnace main circuit are partly regarded as a generalized controlled object, and the below sets up broad sense electric furnace object;

Hydraulic means can be similar to regards a third order PLL joint as, and the transport function form is:

Then: $l_j=\frac{b_0}{a_0{s}^3+a_1{s}^2+a_{2}s}{u}_j,j=a,b,c;---\left(1\right)$

That is: $a_0{l}_j^{\left(3\right)}+a_1{l}_j^{\left(2\right)}+a_2{l}_j^{\left(1\right)}=b_0{u}_j,j=a,b,c;---\left(2\right)$

u _jBe each phase control signal, l _jBe each phase arc length;

Arc Modelling adopts the kohle model, and its form is as being: Z _Arc=R _Arc+ j*X _ArcWherein

R _Arc=l*R _Per.Z _ArcBe arc impedance, R _PerArc resistance value for per unit length; L is arc length, and a gets different values with b, to reflect the different smelting stages;

Furnace transformer is selected the Yd11 connection, utilizes the transformer correlation theory, and in conjunction with electric arc furnaces main circuit electrical system, the effective value that can release the secondary lateral of electric stove transformer electric current is:

Wherein:

Z _Kabc＝(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _b)+

(Z _K/3n-n*Z _a)(Z _K/3n-n*Z _c)+(Z _K/3n-n*Z _b)(Z _K/3n-n*Z _c)

Z _j＝R _d+j*X _d+l _j*R _per+j*(a*l _j*R _per+b*(l _j*R _per) ²)，j＝a，b，c；

Wherein: R _dAnd X _dBe short net impedance and induction reactance, n is the furnace transformer no-load voltage ratio, Be furnace transformer primary side phase voltage, Z _K: the transformer leakage impedance;

Adopt n rank approximation method: ${\mathrm{Ti}}_j^{\left(1\right)}=i_j(k+1)-i_j\left(k\right),$ $T^2{i}_j^{\left(2\right)}=i_j(k+1)-{2i}_j\left(k\right)+i_j(k-1),$ $T^3{i}_j^{\left(3\right)}=i_j(k+1)-{3i}_j\left(k\right)+{3i}_j(k-1)-i_j(k-2),$ J=a, b, c; T is the sampling period; Convolution (1)-(3) obtain the broad sense electric arc furnaces to as the NARMA model being:

i(k+1)＝f[ω(k)，u(k)]+v(k) (4)

Wherein:

f[ω(k)，u(k)]＝[f _a[ω(k)，u(k)]，f _b[ω(k)，u(k)]，f _c[ω(k)，u(k)]] ^T

i(k+1)＝[i _a(k+1)，i _b(k+1)，i _c(k+1)] ^T u(k)＝[u _a(k)，u _b(k)，u _c(k)] ^T

ω(k)＝[i _a(k)，i _a(k-1)，i _a(k-2)，i _b(k)，i _b(k-1)，i _b(k-2)，i _c(k)，i _c(k-1)，i _c(k-2)] ^T

v(k)＝[v _a(k)，v _b(k)，v _c(k)] ^T

F is the nonlinear function vector, f _a, f _b, f _cBe nonlinear function, v (k) is vectorial for disturbing, and establishes || v (k) || and≤v ₀

Second step, adopt LSSVM that broad sense electric furnace object is carried out identification:

The kernel function type of LSSVM adopts gaussian kernel function, and functional form is:

Three-phase current adopts three LSSVM models to carry out identification; Given training dataset, { [ω (k), u (k)], i (k+1) } input data are: [ω (k), u (k)], and the output data are: i (k+1), adopt the learning algorithm of LSSVM, and obtain LSSVM NARMA model and be:

${\hat{i}}_j(k+1)=f[\mathrm{\&omega;}\left(k\right),u\left(k\right)]$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_{t}K\left[\right[\mathrm{\&omega;}\left(k\right),u\left(k\right)],[\mathrm{\&omega;}\left(t\right),u\left(t\right)\left]\right]+b$

$=\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t\exp \left(\frac{-\left|\right|[\mathrm{\&omega;}\left(k\right),u\left(k\right)]-[\mathrm{\&omega;}\left(t\right),u\left(t\right)]\left|\right|}{{2\mathrm{\&sigma;}}^2}\right)+b---\left(5\right)$

j＝a，b，c；

SV expresses support for the number of vector machine, α _tWith b be model parameter;

For NARMA model (4), constantly carry out Taylor at k-1 and launch to obtain approximate model:

i(k+1)＝i(k)+f ₁*Δu(k)+R(k)+v(k) (6)

Wherein: $f_1=\frac{\&PartialD;f[\mathrm{\&omega;}\left(k\right),u\left(k\right)]}{\&PartialD;u\left(k\right)}{|}_{u(.k)=u(k-1),\mathrm{\&omega;}\left(k\right)=\mathrm{\&omega;}(k-1)}$

$f_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]=\frac{{\&PartialD;}^{2}f[\mathrm{\&omega;}(k-1),u\left(k\right)]}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

$R\left(k\right)=\frac{{\left[\mathrm{\&Delta;u}\left(k\right)\right]}^T{f}_2[\mathrm{\&omega;}(k-1),\mathrm{\&zeta;}]\mathrm{\&Delta;u}\left(k\right)}{{\&PartialD;u}^2\left(k\right)}{|}_{u\left(k\right)=\mathrm{\&zeta;}}$

ζ=[ζ _a, ζ _b, ζ _c] ^TAnd u _j(k-1)≤ζ _j≤ u _j(k), j=a, b, c;

By formula (6), ignore R (k) and v (k), the Taylor approximate model that obtains system is:

i(k+1)＝i(k)+f ₁*Δu(k) (7)

The 3rd step, the approximate inverse Internal Model Control Algorithm:

By the output of LSSVM NARMA model (5) derivative being asked in corresponding input obtains

${\hat{f}}_1=\left[\begin{array}{ccc}-\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_a(k-1)\\ -\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_b(k-1)\\ -\frac{u_a(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_b(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)& -\frac{u_c(k-1)}{{\mathrm{\&sigma;}}^2}\underset{t=1}{\overset{\mathrm{SV}}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_t{K}_c(k-1)\end{array}\right]---\left(8\right)$

Wherein: K _j(k-1)=K _j[Q (k-1), Q (t)], j=a, b, c;

And then obtain the computing formula of approximate inverse control algolithm according to formula (7):

u _j(k)＝u _j(k-1)+Δu _j(k) j＝a，b，c； (9)

${\mathrm{\&Delta;u}}_j\left(k\right)={{{\hat{f}}_{1j}}^T({\hat{f}}_1{{\hat{f}}_1}^T+\mathrm{\&alpha;})}^{-1}*(r(k+1)-i\left(k\right))$ When | Δ u _j(k) |≤δ _j

Δ u _j(k)=δ _jSign[Δ u _j(k)] as | Δ u _j(k) |＞δ _j

Wherein: r (k+1) is current setting value, r (k+1)=[r _a(k+1), r _b(k+1), r _c(k+1)] ^T. δ=[δ _a, δ _b, δ _c,] ^T,

δ _jBe arithmetic number,

α=diag (α _a, α _b, α _c), α _jBe very little arithmetic number, sign is sign function;

By prior process as can be known, uncertainty is that inevitably large uncertainty will reduce the performance of control system, weaken uncertainty with the internal mold strategy in real system, when uncertainty exists, according to model and actual output error design control and compensation amount Δ u _c(k), adopt simultaneously robust filter F (z) to weaken uncertainty;

${\mathrm{\&Delta;u}}_{\mathrm{cj}}\left(k\right)=-{{\hat{f}}_{1j}}^T{({\hat{f}}_1{{\hat{f}}_1}^T+\mathrm{\&alpha;})}^{-1}*F(\&CenterDot;z)*(R\left(k\right)+v\left(k\right))$ When | Δ u _Cj(k) |≤δ _Cj(10)

Δ u _Cj(k)=δ _CjSign[Δ u _Cj(k)], as | Δ u _Cj(k) |＞δ _Cj

J=a in the formula, b, c; δ _c=[δ _Ca, δ _Cb, δ _Cc], δ _CjBe arithmetic number, sign is sign function, and R (k)+v (k) uses

Replace;

To sum up must be with the approximate inverse Internal Model Control Algorithm of compensation:

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) (11)

Described excess current control algorithm is: the material that prevents from collapsing waits and causes instantaneous short circuit electrode mistuning joint to occur, whenever, design mutually a window integrator, when the current integration value surpasses the maximum limit definite value, the work of excess current control algorithm, so that excess current is worth larger, electrode speed raises faster.Like this, by promoting fast single rather than a plurality of relevant electrodes, correct overcurrent condition, the vibration that brings to avoid the electrode coupling impact is regulated, and avoids the overload of furnace transformer and electrode; Specific algorithm is:

$s_{\mathrm{nj}}=\underset{k-n-1}{\overset{k}{\mathrm{\&Sigma;}}}{i}_j\left(k\right)$

u _j(k)＝K _tS _nj S _nj≥L _max (12)

u _j(k)＝u _j(k-1)+Δu _j(k)+Δu _cj(k) S _nj＜L _max

In the formula, n is window integrator length, K _tBe overcurrent controller scale-up factor, L _MaxBe current integration maximum limit definite value;

Described automatic arc striking algorithm is: when automatically moving, at first make three-phase electrode automatically drop to starting the arc position, then automatically adjust; Control method is: during automatic arc striking, A phase electrode descends automatically, and T rises after the A phase voltage drops to predetermined voltage _aStop second, and then B phase electrode descends automatically, and T rises after the B phase voltage drops to predetermined voltage _bStop second, and last C phase electrode descends automatically, and T rises after the C phase voltage drops to predetermined voltage _cStop T second _a, T _b, T _cBy electrode speed of feed model T _j=L/V _jObtain, j=a, b, c, in the formula, L is for preparing striking-distance; Vi is the electrode ascending velocity.

Images