CN104699039A - Expert control method for burning double-chamber lime furnace - Google Patents

Abstract

The invention discloses an expert control method for burning a double-chamber lime furnace. The method comprises the following steps: determining operation quantity and controlled quantity, collecting and classifying data, identifying models, obtaining prediction models, and performing optimal solution; the method can solve the problem that coupling among variables, system nonlinearity, time lag and temperature in the furnace cannot be accurately measured, the excellent control effect can be realized, the coal injection quantity of the controlled quantity and the aperture of the furnace tail blower are automatically adjusted according to the set values of NOx and O2 contents of the controlled quantity for burning the double-chamber lime furnace, so that the actual value of the controlled quantity can quickly follow up the set value, the operating quality during the burning process of the double-chamber lime furnace is increased, the operation during the burning process is relatively steady, reasonable and accurate, and the labor intensity of workers is also reduced.

Description

A kind of LIME DOUBLE thorax kiln calcining expert control method

Technical field

The invention belongs to the advanced control method technical field of LIME DOUBLE production run, relate to a kind of LIME DOUBLE thorax kiln calcining expert control method.

Background technology

The calcining of LIME DOUBLE thorax kiln is a very typical very important link in lime manufacturing process, directly affects the quality of lime, production capacity and production cost to its operational quality.The quality of LIME DOUBLE thorax kiln calcination operation is decided by again the control method to it to a great extent, so the advanced control method of LIME DOUBLE thorax kiln calcining is also just more and more subject to concern both domestic and external and attention.

The controlling difficulties of LIME DOUBLE thorax kiln calcining be mainly its have complicated Coupled Variable, non-linear and time stickiness, cause traditional experience to regulate method, PID to regulate the control method such as method, fuzzy control to be difficult to obtain good, stable control effects to it.Particularly country makes the research of LIME DOUBLE thorax kiln calcination process seem more and more important to the requirement of the aspect such as economic restructuring, energy-saving and emission-reduction.Numerous domestic and international technique, automatic control expert have done a large amount of research work to this, propose many advanced control theories, also achieve some challenging progress.But in view of restricted to control object of the complicacy of LIME DOUBLE thorax kiln calcination process and advanced control theory, make the control algolithm of some advanced persons and control program in LIME DOUBLE thorax kiln calcination process, not be able to fine enforcement.

Summary of the invention

The present invention in order to overcome the defect of prior art, devise one can solve coupling between variable, mission nonlinear, time stickiness and kiln temperature can not Measurement accuracy problem, and the LIME DOUBLE thorax kiln of the effect that can be well controlled calcining expert control method.

The concrete technical scheme that the present invention takes is: a kind of LIME DOUBLE thorax kiln calcining expert control method, and key is: described method comprises the following steps:

A, determination operation amount and controlled volume: using the injecting coal quantity of LIME DOUBLE thorax kiln calcination system and kiln tail blower fan baffle opening as operational ton, NO in kiln _xcontent and O ₂content is as controlled volume;

B, data acquisition and classification: the operational ton data of the on-the-spot LIME DOUBLE thorax kiln calcination process determined in acquisition step a and controlled volume data, and institute's image data is divided into steady state data and dynamic data two class;

C, Model Distinguish: Nonlinear Steady Model Distinguish is carried out to the steady state data obtained in step b, set up the Nonlinear Steady model of LIME DOUBLE thorax kiln calcination process, linear dynamic model identification is carried out to the dynamic data obtained in step b, sets up the ergodic ARX model of linear dynamic autoregression of LIME DOUBLE thorax kiln calcination process;

D, obtain forecast model: the Nonlinear Steady model obtain identification in step c and the ergodic ARX model parallel connection of linear dynamic autoregression organically combine, and adopt iteration recurrence method, obtain the forecast model of LIME DOUBLE thorax kiln calcination process;

E, optimum solve: set up objective function according to the forecast model obtained in steps d, carry out optimum and solve, obtain the predicted value of LIME DOUBLE thorax kiln controlled quentity controlled variable accurately to objective function.

Described step b uses PLC module to gather the operational ton data in LIME DOUBLE thorax kiln calcination process and controlled volume data, according to step response characteristic, the execute-in-place amount data gathered and controlled volume data are classified again, be divided into steady state data and dynamic data two class.

Nonlinear Steady Model Distinguish described in step c is the discrimination method utilizing fuzzy neural network, carries out identification to the steady state data screened.

Linear dynamic model identification described in step c utilizes least square method, carries out identification to the dynamic data screened.

Described step e adopts quadratic programming SQP to carry out optimum to objective function to solve.

The invention has the beneficial effects as follows: (1) selects NO _xcontent and O ₂content is as two output quantities of LIME DOUBLE thorax kiln calcining predictive controller, and select kiln hood injecting coal quantity and kiln tail blower fan baffle opening as two input quantity, the predictive controllers that employing two enters scene 2 can solve the coupled relation between these four variablees; (2) model of LIME DOUBLE thorax kiln calcination process is divided into Nonlinear Steady model and linear dynamic model, and two models combinations are carried out PREDICTIVE CONTROL to two thorax kiln calcining, be embodied in forecast model by system time lags again, the method can solve non-linear in calcination process and time lag characteristic; (3) forecast Control Algorithm of the present invention can according to LIME DOUBLE thorax kiln calcining controlled volume NO _xcontent and O ₂the setting value of content, automatically its controlled quentity controlled variable injecting coal quantity and kiln tail blower fan baffle opening is regulated, the actual value of controlled volume can be followed up its setting value rapidly, improve the operational quality of LIME DOUBLE thorax kiln calcination process, make the operation of calcination process more steadily, rationally and accurately, also reduce the labour intensity of workman simultaneously.

Accompanying drawing explanation

Fig. 1 is schematic flow sheet of the present invention.

Fig. 2 is the built-up pattern schematic diagram composed in parallel by Nonlinear Steady model and linear dynamic model in the present invention.

Embodiment

A kind of LIME DOUBLE thorax kiln calcining expert control method, comprises the following steps:

A, determination operation amount and controlled volume: LIME DOUBLE thorax kiln calcination process can be divided into material road and gas circuit two circuits, the trend on material road is: lime raw material enter kiln thorax preheating section under feeding system effect, enter calcining section more under gravity to calcine, finally by output finished product after cooling section; The trend of gas circuit is just in time contrary in the effect Xia Yuliao road of kiln tail exhaust fan trend; In above-mentioned technological process, the quality of two thorax kiln calcining directly affects quality and the production capacity of lime, and the quality of calcining mainly depends on again burning of coal situation in the accurate control of two thorax kiln kiln temperature and kiln.Find NO in the height of kiln temperature and kiln after deliberation _xcontent there is a kind of fixed relationship, the higher NO of temperature _xcontent is higher, and the flowing velocity of gas in kiln is very fast, so with kiln tail NO _xcontent can reflect kiln temperature rapidly, simultaneously also find kiln tail O ₂content can reflect burning of coal situation in kiln discharge, so using injecting coal quantity and kiln tail blower fan baffle opening as operational ton, by NO in kiln _xcontent and O ₂content is as controlled volume;

B, data acquisition and classification: to NO in LIME DOUBLE thorax kiln calcination process _xcontent, O ₂content, injecting coal quantity and kiln tail blower fan baffle opening four variablees carry out on-site data gathering by DCS system, sampling period is 60s, gather 20000 groups of data altogether, institute's image data is divided into dynamically, stable state two kinds of data, partitioning standards is: when in operational ton injecting coal quantity and kiln tail blower fan baffle opening, Spline smoothing occurs for any one, controlled volume NO _xcontent and O ₂content elects dynamic data from the step moment as to data when finally all reaching steady state (SS); Data before controlled volume occurs from steady state (SS) to next step elect steady state data as, after classification, obtain dynamic data 7341 groups, steady state data 12659 groups;

C, Model Distinguish: be a very important link to the Model Distinguish of kiln tail blower fan calcination process, it is divided into Nonlinear Steady Model Distinguish and linear dynamic model identification two parts, first by NO in kiln before Model Distinguish _xcontent and O ₂content is defined as two output quantity y of model ₁and y ₂, injecting coal quantity and kiln tail blower fan baffle opening are defined as input quantity u ₁and u ₂,

Nonlinear Steady Model Distinguish is the discrimination method utilizing fuzzy neural network, carries out identification to the steady state data screened, and sets up the Nonlinear Steady model of LIME DOUBLE thorax kiln calcination process, shown in (1):

y _s＝f ₂[W _2,1·f ₁(W _1,1·u _s+θ ₁)+θ ₂] (1)

In formula, that systematic steady state exports, systematic steady state input, f ₁, f ₂the transport function of hidden layer, output layer respectively, W _1,1, W _2,1the weights of hidden layer, output layer respectively, θ ₁, θ ₂the threshold values of hidden layer, output layer respectively;

Linear dynamic model identification is the discrimination method utilizing least square method, first the dynamic data screened is processed, deduct the input and output value in a upper moment by the input and output value of current time respectively and obtain dynamic change incremental data, again Model Distinguish is carried out to increments of change data, set up the ergodic ARX model of linear dynamic autoregression of LIME DOUBLE thorax kiln calcination process, shown in (2):

${\mathrm{\Δy}}_{\mathrm{\λ}}\left(k\right)=A_1^{\left(\mathrm{\λ}\right)}\mathrm{\Δy}(k-1)+A_2^{\left(\mathrm{\λ}\right)}\mathrm{\Δy}(k-2)+B_1^{\left(\mathrm{\λ}\right)}\mathrm{\Δu}(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-1)+B_2^{\left(\mathrm{\λ}\right)}\mathrm{\Δu}(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-2)---\left(2\right)$

In formula, Δ y (k)=y (k)-y (k-1), Δ u (k)=u (k)-u (k-1), y (k)=[y ₁(k) y ₂(k) L y _ny(k)] ^t, u (k-τ ^(λ))=[u ₁(k-τ _{λ 1}) u ₂(k-τ _{λ 2}) L u _nu(k-τ _{λ nu})] ^toutput vector and the input vector of system respectively, A ^(λ)represent that the λ getting matrix A is capable, λ=1,2, L, ny, A ₁, A ₂∈ R ^{ny × ny}, B ₁, B ₂∈ R ^{ny × nu}, ny is controlled volume number ny=2, nu is controlled quentity controlled variable number nu=2;

D, obtain forecast model: on the basis obtaining LIME DOUBLE thorax kiln calcination process Nonlinear Steady model and linear dynamic model, derive the forecast model of calcination process, first steady-state model and dynamic model are organically combined by the steady-state gain K of system, as shown in Figure 1, form a built-up pattern in parallel, linear dynamic autoregression ergodic ARX model is now transformed to the ergodic ARX model of linear dynamic autoregression containing steady-state gain K parameter, shown in (3):

$\begin{array}{c}{\mathrm{\Δy}}_{\mathrm{\λ}}\left(k\right)=A_1^{\left(\mathrm{\λ}\right)}\mathrm{\Δy}(k-1)+A_2^{\left(\mathrm{\λ}\right)}\mathrm{\Δy}(k-2)+B_{11}^{\left(\mathrm{\λ}\right)}\mathrm{\Δu}(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-1)+B_{12}^{\left(\mathrm{\λ}\right)}\mathrm{\Δu}(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-2)\\ +B_{21}^{\left(\mathrm{\λ}\right)}{\mathrm{\Δu}}^2(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-1)+B_{22}^{\left(\mathrm{\λ}\right)}{\mathrm{\Δu}}^2(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)}-2)\end{array}---\left(3\right)$

In formula,

B ₁₁＝A ₀g[B ₁'./(B ₁'+B' ₂).*K ^c]，B ₁₂＝A ₀g[B' ₂./(B ₁'+B' ₂).*K ^c]

B ₂₁＝A ₀g[B ₁'./(B ₁'+B' ₂).*K']，B ₂₂＝A ₀g[B' ₂./(B ₁'+B' ₂).*K']

K'＝(K ⁿ-K ^c)./(u _s(k+1)-u _s(k))

${\mathrm{\Δu}}^2(k-{\mathrm{\τ}}^{\left(\mathrm{\λ}\right)})={\left[{\mathrm{\Δu}}_1^2(k-{\mathrm{\τ}}_{\mathrm{\λ}1}){\mathrm{\Δu}}_2^2(k-{\mathrm{\τ}}_{\mathrm{\λ}2})L{\mathrm{\Δu}}_{\mathrm{nu}}^2(k-{\mathrm{\τ}}_{\mathrm{\λnu}})\right]}^T,$ B ₁' and B ₂' be obtained, K by transformation of coefficient in formula (2) ^c, K ⁿstable state input vector u respectively _s(k), u _s(k+1) corresponding dynamic gain matrix, the value of other symbol is identical with the value of formula (2), B ₁.*B ₂represent B ₁and B ₂the element multiplication of middle correspondence position, K./u _srepresent the element of every a line in K and vectorial u _sthe element of middle correspondence position is divided by;

The forecast model expression formula containing time lag and steady-state gain K is obtained, shown in (4) by formula (3) recursion:

In formula, g ₁₁, G ₁₂∈ R ^{(nyP) × (nuM)}; Δ U (k), Δ U ²(k) ∈ R ^(nu. ^{m) × 1}, F ₁, F ₂∈ R ^{(nyP) × ny};

$G_{21},{G_{22\∈R}}^{(\mathrm{ny}\·P)\×[\mathrm{nu}\·(\mathrm{ny}+\underset{i=1}{\overset{\mathrm{ny}}{\mathrm{\Σ}}}{d}_i)};{\mathrm{\ΔU}}_2\left(k\right),{{{\mathrm{\ΔU}}_2^2\left(k\right)}_{\∈R}}^{[\mathrm{mu}\·(\mathrm{ny}+\underset{i=1}{\overset{\mathrm{ny}}{\mathrm{\Σ}}}{d}_i)]\×1};$ Δ y (k), Δ y (k-1) ∈ R ^{ny × 1}; d _λ=max{ τ _{λ 1}, τ _{λ 2}, L, τ _{λ nu}(λ=1,2, L, ny), wherein P is prediction step number, and M is for controlling step number, τ ₁₂for u ₂to y ₁time lag step number, for the Increment Matrix of controlled volume predicted value, Δ y (k), Δ y (k-1) is the current step k of controlled volume and the Increment Matrix in previous step k-1 moment, Δ U (k), Δ U ²k () is the Increment Matrix of controlled quentity controlled variable predicted value, Δ U ₂(k), for the Increment Matrix in the controlled quentity controlled variable time lag time period in the past, G ₁₁, G ₁₂, G ₂₁, G ₂₂, F ₁, F ₂include steady-state gain K and dynamic ARX model parameter A ₁, A ₂, B ₁, B ₂parameter matrix;

E, optimum solve: process objective function being asked to optimum solution, are also the committed steps of LIME DOUBLE thorax kiln calcining PREDICTIVE CONTROL, the objective function that the predicted value of foundation controlled volume and controlled quentity controlled variable and history value build and constraint function, shown in (5):

$\min J\left(k\right)={\left|\right|{\hat{Y}}_c-Y_r\left(k\right)\left|\right|}_Q^2+{\left|\right|\mathrm{\ΔU}\left(k\right)\left|\right|}_R^2$

s.tΔu _min≤Δu(k+i)≤Δu _max

u _min≤u(k+i)≤u _max,(i＝0,1,L,M-1)

y _min≤y(k+j)≤y _max,(j＝1,2,L,P) (5)

In formula, Q=qI, R=rI are the weighting matrix of output error and controlling increment respectively, and I is unit matrix, the revised predicted value matrix of controlled volume, Y _rk () is the reference locus matrix of controlled volume, Δ u is controlled quentity controlled variable Increment Matrix, u and y is controlled quentity controlled variable matrix and controlled volume matrix, Δ u _min, Δ u _max, u _min, u _max, y _min, y _maxfor the bound binding occurrence to dependent variable;

For the objective function of formula (5) non-linear belt constraint, the present invention adopts quadratic programming SQP to solve its optimum, obtain controlling the control variable quantity in step number M, quadratic programming SQP solving method target problem is converted to a series of quadratic programming subproblem to solve, and has good effect to solving non-linear belt restricted problem.

The data that performance index compare are studied as shown in table 1 with adopting similar pair of thorax kiln of existing general control method:

Table 1

Claims (5)

1. a LIME DOUBLE thorax kiln calcining expert control method, is characterized in that: described method comprises the following steps:

A, determination operation amount and controlled volume: using the injecting coal quantity of LIME DOUBLE thorax kiln calcination system and kiln tail blower fan baffle opening as operational ton, NO in kiln _xcontent and O ₂content is as controlled volume;

B, data acquisition and classification: the operational ton data of the on-the-spot LIME DOUBLE thorax kiln calcination process determined in acquisition step a and controlled volume data, and institute's image data is divided into steady state data and dynamic data two class;

C, Model Distinguish: Nonlinear Steady Model Distinguish is carried out to the steady state data obtained in step b, set up the Nonlinear Steady model of LIME DOUBLE thorax kiln calcination process, linear dynamic model identification is carried out to the dynamic data obtained in step b, sets up the ergodic ARX model of linear dynamic autoregression of LIME DOUBLE thorax kiln calcination process;

D, obtain forecast model: the Nonlinear Steady model obtain identification in step c and the ergodic ARX model parallel connection of linear dynamic autoregression organically combine, and adopt iteration recurrence method, obtain the forecast model of LIME DOUBLE thorax kiln calcination process;

E, optimum solve: set up objective function according to the forecast model obtained in steps d, carry out optimum and solve, obtain the predicted value of LIME DOUBLE thorax kiln controlled quentity controlled variable accurately to objective function.

2. a kind of LIME DOUBLE thorax kiln calcining expert control method according to claim 1, it is characterized in that: described step b uses PLC module to gather the operational ton data in LIME DOUBLE thorax kiln calcination process and controlled volume data, according to step response characteristic, the execute-in-place amount data gathered and controlled volume data are classified again, be divided into steady state data and dynamic data two class.

3. a kind of LIME DOUBLE thorax kiln calcining expert control method according to claim 1, is characterized in that: the Nonlinear Steady Model Distinguish described in step c is the discrimination method utilizing fuzzy neural network, carries out identification to the steady state data screened.

4. a kind of LIME DOUBLE thorax kiln calcining expert control method according to claim 1, is characterized in that: the linear dynamic model identification described in step c utilizes least square method, carries out identification to the dynamic data screened.

5. a kind of LIME DOUBLE thorax kiln calcining expert control method according to claim 1, is characterized in that: described step e adopts quadratic programming SQP to carry out optimum to objective function to solve.