CN102841540A - MMPC-based supercritical unit coordination and control method - Google Patents

Abstract

The invention discloses an MMPC (Multiple Model Predictive Control)-based supercritical unit coordination and control method which comprises the following steps: pre-arranging a plurality of local predictive models according to the nonlinear working condition response of a supercritical unit; selecting a load parameter to serve as a control variable through a model identification method, and obtaining the output increments of all local predictive models through an interpolation formula; and in each control period, piling up the output increments of all local predictive models through a DMC (Dynamic Matrix Control) prediction control algorithm so as to obtain an actual output increment which is used for correcting a feedforward passage. According to the invention, the MMPC method is adopted for the coordination and control of the supercritical unit; a selection method of the control variable, a process variable and a perturbation variable for prediction control is proposed; and a specific construction scheme of the MMPC in a coordination and control system is also provided. The method has the advantages of advancement, strong practicability, good robustness, and quick and stable control effect, as a result, an ideal control effect is obtained.

Description

Supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL

Technical field

The present invention relates to industrial process control technology field, particularly relate to supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL.

Background technology

The automatic control of Power Plant Thermal Process system is one of hot issue of control circle research always; Along with rapid development of social economy; The working condition of production system is complicated day by day; Require to improve day by day, control system often demonstrates characteristics such as multivariate, strong coupling, operating mode scope are wide, the comprehensive requirement height of control performance, makes the research of this problem become complicated more.Solve challenge in daily life, people usually unconsciously utilization decompose compositional rule, promptly the big problem of complicacy can be decomposed into one group of simple minor issue, with each minor issue carry out certain synthetic push away the separating of former problem.Control problem for complication system; Above-mentioned rule has reference property equally; With the complex nonlinear PROBLEM DECOMPOSITION is some simple linear problems, finds the solution to obtain good modeling and control effect to each linear problem then, and this multi-model control strategy just occurred; Just the someone proposed through the precision of prediction of model and the method for robustness are improved in several model phases Calais as far back as 1969; People have generally believed that it is a kind of simple and effective way that possibly obtain better to predict the outcome that several models are combined now, also emerge in an endless stream based on the multi-model modeling method, like T-S Fuzzy Multiple Model, partial model network, thresholding autoregression or the like.The multi-model controller is suggested the seventies; Be mainly used in and solve control problem, promptly controlled device is set up a plurality of models, cover its parameter uncertainty with parameter uncertainty system; Utilize output error to ask for weights, ask for controller through the mode of weighted sum then.

In the middle of conventional art, coordinate control for the power plant boiler tradition, adopt MPC (Model Predictive Control, Model Predictive Control) technology mostly, but the control effect of practical application is unsatisfactory.Fired power generating unit load variations will be with boiler and steam turbine by a Steady-State Control to another stable state; Existing ripe MPC technology adopts linear model more; And the non-linear process modeling all can not adapt to the significantly variation of load with control; Normal load and the bigger situation of pressure divergence of occurring is very difficult in the application of real process control.

Summary of the invention

Based on this, being necessary provides a kind of supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL to the problems referred to above, can the multi-model forecast Control Algorithm be used for the coordination control of ultra supercritical unit, obtains the control effect of quick and stable.

A kind of supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL comprises:

Non-linear operating mode response according to supercritical unit is provided with preset several local prediction models;

Through the Model Identification method, choose load parameter as control variable, obtain the output increment of each local prediction model through interpolation formula;

In each control cycle, superpose through the output increment of DMC predictive control algorithm said each local prediction model, obtain actual output increment feedforward path is proofreaied and correct.

Embodiment of the present invention has following beneficial effect:

The present invention combines multi-model with PREDICTIVE CONTROL, solve the ultra supercritical fired power generating unit and coordinate control problem.Multi-model PREDICTIVE CONTROL (MMPC) is a kind of citation form of nonlinear prediction control, adopts a plurality of linearizing partial models to describe same non-linear object.MMPC control is to the partial model CONTROLLER DESIGN, and the employing method of interpolation is carried out the model switching.Confirmed reasonable number and the division methods of control, proposed the concrete scheme that PREDICTIVE CONTROL is implemented in coordinated control system with the choosing method and the multi-model PREDICTIVE CONTROL of control variable, process variable and disturbance variable with model.The present invention has advance, and is practical, and robustness is good.This invention control effect quick and stable can obtain ideal control effect.

Description of drawings

Fig. 1 is the process flow diagram that the present invention is based on the supercritical unit control method for coordinating of multi-model PREDICTIVE CONTROL;

Fig. 2 is the synoptic diagram of the step response of input and output of the present invention;

Fig. 3 is the synoptic diagram of different load section multi-model switching principle of the present invention;

Fig. 4 is the embodiment synoptic diagram that the present invention is based on the supercritical unit control method for coordinating of multi-model PREDICTIVE CONTROL.

Embodiment

For making the object of the invention, technical scheme and advantage clearer, the present invention is done to describe in detail further below in conjunction with accompanying drawing.

Got into since the eighties; People begin further research based on the multi-model controller of stablizing under the meaning; And attempt multi-model is combined with various conventional controls; Like multi-model Adaptive Control, multi-model fuzzy control, multi-model PREDICTIVE CONTROL or the like, also obtained good effect in practice.Research through a large amount of shows, it is found that and adopts the multi-model strategy to make NLS that model simplification, transparent and controller arranged, and is convenient to systematic analysis; Compare with other non-linear global policies, its computational complexity reduces greatly; Model and controller architecture are more suitable for online flexibly adjustment learning algorithm; Adopt this modeling structure to be convenient on higher level, add qualitative information; The more important thing is that people are the main factors of impelling this scheme gained popularity and bearing fruit for the familiar and grasp degree of linear model and controller.Just because of above-mentioned many characteristics, multi-model control receives increasing concern, and is used widely in fields such as process control, aviation, medical science.In recent years; The main scholarly journal of international automation field such as IEEETrans-AC, Automatica and important international conference such as the U.S. control meeting (ACC), IEEE control and decision making meeting (CDC) etc. and have delivered many papers about multi-model control; Some scholars also sum up forefathers' work; International publication Int.J.Control has published the multi-model process monograph 1997 7/8 phases, makes the status on multi-model theory control circle in the world obtain again certainly.

The present invention controls the coordination that the multi-model forecast Control Algorithm is used for the ultra supercritical unit, has proposed the concrete scheme that PREDICTIVE CONTROL is implemented in coordinated control system with the choosing method and the multi-model PREDICTIVE CONTROL of control variable, process variable and disturbance variable.

Fig. 1 is the process flow diagram that the present invention is based on the supercritical unit control method for coordinating of multi-model PREDICTIVE CONTROL, comprising:

S101: the non-linear operating mode response according to supercritical unit is provided with preset several local prediction models;

S102: through the Model Identification method, choose load parameter, obtain the output increment of each local prediction model through interpolation formula as control variable;

S103: in each control cycle, superpose, obtain actual output increment feedforward path is proofreaied and correct through the output increment of DMC predictive control algorithm with said each local prediction model.

The present invention confirms the reasonable number of control with the local prediction model according to the non-linear strong characteristics of ultra supercritical unit, uses classical DMC predictive control algorithm to find the solution controller output to the local prediction controller.And on the basis of coordinating control based on bottom DCS, implement the PREDICTIVE CONTROL of multi-model, and can seamlessly transit the control output in period like this, effectively improve the load and the pressure response characteristic of unit, adapt to large-scale working conditions change.

At first setting up the unit model of one group of RP according to power of the assembling unit L, serves as the scheduling variable with load L then, and the interpolation rule can be defined as:

MV(k)=ω(k-1)MV(k-1)+ω(k+1)MV(k+1)

$\mathrm{\ω}(k-1)=1-\frac{L\left(k\right)-L(k-1)}{L(k+1)-L(k-1)}$

$\mathrm{\ω}(k-1)=1-\frac{L(k+1)-L\left(k\right)}{L(k+1)-L(k-1)}---\left(1\right)$

Wherein L (k) is current unit real power; L (k-1) is the power of the assembling unit less than the nearest modeling point of L (k); L (k+1) is the power of the assembling unit greater than the nearest modeling point of L (k), and MV (k) is current controlled quentity controlled variable output, and MV (k-1) locates the controlled quentity controlled variable that Model Calculation goes out for the power of the assembling unit at L (k-1); MV (k+1) locates the controlled quentity controlled variable that Model Calculation goes out for the power of the assembling unit at L (k+1), and ω (k-1) is MV (k-1) and the shared ratio of MV (k+1) among the current controlled quentity controlled variable MV (k).

Fig. 2 is the synoptic diagram of the step response of input and output of the present invention.

Said load parameter comprises: valve opening, fuel quantity instruction, Total Feedwater Flow.

In certain operating mode segment limit; The model of unit can be approximated to be linear model; Can set up corresponding relation through applying step disturbance like the step response of the multiple-input and multiple-output of Fig. 2, when the curve among the figure is the unit's of being input as increment, the response curve of corresponding output.With the sampling period is length; Get the point value of selected prediction step length; Can form the vector of describing the dynamic response characteristic between the input and output, these vectors are arranged in together, just form the dynamic response matrix of prediction total system input and output dynamic perfromances.Suc as formula (2):

Y _t+1|t＝Y _t+1|t1+AΔU _t （2）

Y _T+1|tBe t model prediction constantly output, Y _T+1|t-1For t zero imports free response vector, Δ U constantly _tT is the control increment vector constantly, and A is a dynamic matrix.

T is dynamic matrix constantly:

Inscribe step response FIR (finite impulse response (FIR)) model of system's output during t to i control input:

${[a_i^{\left(1\right)},a_i^{\left(2\right)},...,a_i^{\left(N\right)},...,a_i^{\left(P\right)},...,a_i^{\left(N\right)}]}^T$

Make the minimum deviation between prediction output and the output setting value, the PREDICTIVE CONTROL controller of single model is output as:

$\underset{\mathrm{\ΔU}}{\min }J\left(t\right)={\left|\right|W\left(t\right)-Y_M\left(t\right)\left|\right|}_Q^2+{\left|\right|\mathrm{\ΔU}\left(t\right)\left|\right|}_R^2---\left(5\right)$

Wherein,

The mathematics implication of weighted norm square is:

System's output setting value vector: W (t)=[w (t+1), w (t+2) ..., w (t+P)] ^T

Error weight matrix: Q=diag [Q ⁽¹⁾, Q ⁽²⁾..., Q ^(P)];

T is error power constantly: $Q^{\left(t\right)}=\mathrm{Diag}[q_1^{\left(t\right)},q_2^{\left(t\right)},...,q_m^{\left(t\right)}];$

Control matrix: R=diag [R ⁽¹⁾, R ⁽²⁾..., R ^(M)];

K is control constantly: $R^{\left(t\right)}=\mathrm{Diag}[r_1^{\left(t\right)},r_2^{\left(t\right)},...,r_n^{\left(t\right)}];$

Rolling optimization strategy under the nothing constraint of inscribing in the time of can obtaining k thus:

ΔU(k)=(ATQA+R) ^-1A ^TQ[W(k)-Y ₀(k)](6)

Error weight matrix Q has characterized in a following P prediction step the error control degree, and control matrix R has then characterized the degree of restraint to control increment.On the basis of (6) formula, the control law u (t) that DMC inscribes when system is applied t:

Δu(k)=Δu(k|k)=d ^T[W(k)-Y ₀(k)]； (7)

u(k)=u(k-1)+Δu(k) (8)

Wherein, d=[c ^T(A ^TQA+R) ^-1A ^TQ] ^-1, c=[I, 0 ..., 0] ^T

Setting is based on the local prediction model of 50% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 70% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 95% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 100% boiler maximum continuous rating load section.

Obtain the controlled quentity controlled variable of local prediction controller through predictive control algorithm after; Set up the unit input and coordinate 30% of control; 50%, 75% and the model of four of 100%MCR load section, the switching between the model adopts interpolation method to carry out; Guarantee the accuracy of MPC control, the structure control of different load section multi-model is as shown in Figure 3.

Fig. 3 is the synoptic diagram of different load section multi-model switching principle of the present invention.R is the input setting value among the figure, and Y is output, measures after the correction of reference locus for the input setting value, is generally first-order filtering, makes the actual set value with the close setting value of certain speed R, and big fluctuation appears in anti-locking system.Forecast model and rolling optimization are the link of fundamental forecasting control algolithm, obtain following output of system through forecast model, obtain next optimum control amount constantly through rolling optimization.CV among the figure (Controlled Variable) is a controlled variable; Some key parameter for controlled device; In this patent the preceding pressure of machine, unit load and the separator outlet temperature etc. of unit; MV (Manipulated Variable) is a performance variable, is to make controlled variable maintain the variable of the required adjusting of setting value, and in this patent fuel quantity, main steam speed governing valve opening and the confluent of unit; DV (Disturbed Variable) is a disturbance variable, change of this patent middle finger ature of coal and network load fluctuation.

Fig. 4 is the embodiment synoptic diagram that the present invention is based on the supercritical unit control method for coordinating of multi-model PREDICTIVE CONTROL.Below in conjunction with Fig. 4 embodiment of the present invention is described.

Step 1: the early-stage preparations of controller

In the early-stage preparations of multi-model predictive controller, need be familiar with the technological process of process object, and then accomplish the selection of controller input/output variable, accomplish like the corresponding figure of the input and output of Fig. 2.Said load parameter comprises: valve opening, fuel quantity instruction, Total Feedwater Flow.

Step 2: the establishment of multi-model zone segmentation and switching law

According to the non-linear operating mode response of supercritical unit, in the scope of operation operating mode, the non-linear dynamic characteristic piece-wise linearization of object is handled, preset several local prediction models are set.Guarantee after the segmentation, in each sectional area, can be similar to process is treated to linear process and does not influence the quiet run of system.After sectional area is clear and definite, can establish switching law based on formula (1) and Fig. 2.Preferably,

Setting is based on the local prediction model of 30% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 50% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 75% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 100% boiler maximum continuous rating load section.

Step 3: the test of process object data

Related numerous sensor of multi-model predictive controller and topworks, like the related transmitter of input/output variable or valve if fault all to repair, to guarantee the smooth input of multi-model predictive controller.If confirm that multi-model predictive controller input related sensor and topworks are all normal, then can further carry out the test of process object.Test is through each input variable is carried out upset test, record simultaneously, the data of gatherer process; Concrete condition according to sampling period and image data is carried out filtering to data; Remove data noise, set up input and output step response corresponding relation, if the model curve of obviously running counter to operational characteristic is arranged like Fig. 2; Then need look into test again, up to measuring satisfied model.

Step 4: the emulation and the adjustment of setting up the controller configuration file and carrying out the controller off-line

Through the model of identification, utilize the realistic model of unit to carry out the emulation and the parameter adjustment of controller, thus the variation of bound of test disturbance variable, CV setting value, MV and CV or the like.Can further adjust the performance that parameter obtains expecting after the performance of the controller of assessment.

Step 5: the online trial run of multi-model predictive controller

Whether normally multi-model predictive controller On-line Control program comes check program operation, the accuracy of testing model simultaneously at first with pretest operational mode operation one to two day.Under this pattern, controller will be accomplished various computings, but the output of controller is not added on the controlled device.All MV bounds will be fixed on the very approaching scope of current setting value in, simultaneously controller can only have very little action.Start the multi-model predictive controller, control variable is applied on the object,, adjust again in case of necessity through the control performance of evaluation and test controller.

Step 6: the maintenance of multi-model predictive controller

For the multi-model predictive controller, need to safeguard the optimum of guaranteeing performance.Can to confirm them in allowed limits, remove unessential bound simultaneously through detecting MV and CV operation bound,

, said load parameter is adjusted during in the amplitude of variation of Preset Time internal loading parameter according to the local preset model under current greater than threshold value.

When the non-shutdown peak regulation of unit; Continuous adjustment by a relatively large margin takes place in the load meeting, at this moment is responsible for to the supervision level module of load changing rate, when the internal loading variation surpassed a certain setting value in three minutes (like 9MW/min); Unit design Maximum speed limit 12MW/min; Consider by its 75% speed), this supervision grade module can trigger and according to affiliated load section coal amount, the water yield etc. revised in advance, reacts the toning problem that possibly cause slowly at this moment thereby remedy owing to switch scheduling logic.

Surpass preset crossing the border during time span when getting into another operating mode, carry out the switching scheduling between each local prediction model.

In certain operating mode segment limit, the model of unit can be approximated to be linear model, can set up the input and output step response matrix relationship of similar Fig. 2 through applying step disturbance.Set up the model of four the load sections of 30%, 50%, 75% and 100%MCR after the unit input is coordinated to control, the switching between the model adopts interpolation method to carry out, and guarantees the accuracy of the PREDICTIVE CONTROL control of multi-model.Multi-model structure control of different load section such as needs adopt output filtering and switch scheduling.When output filtering is used for hand/automatic switchover and model blocked operation; Only when getting into new operating mode above a time set value (such as 10 minutes); Just can carry out model switches; Thereby avoided having strengthened the stability of system because operating mode is unstable or the frequent model of the controller that when two operating mode critical zones work, possibly cause switches.

The above embodiment has only expressed several kinds of embodiments of the present invention, and it describes comparatively concrete and detailed, but can not therefore be interpreted as the restriction to claim of the present invention.Should be pointed out that for the person of ordinary skill of the art under the prerequisite that does not break away from the present invention's design, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with accompanying claims.

Claims (5)

1. the supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL is characterized in that, comprising:

Non-linear operating mode response according to supercritical unit is provided with preset several local prediction models;

Through the Model Identification method, choose load parameter as control variable, obtain the output increment of each local prediction model through interpolation formula;

In each control cycle, superpose through the output increment of DMC predictive control algorithm said each local prediction model, obtain actual output increment feedforward path is proofreaied and correct.

2. the supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL according to claim 1 is characterized in that said load parameter comprises: valve opening, fuel quantity instruction, Total Feedwater Flow.

3. the supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL according to claim 1 and 2 is characterized in that, the step that preset several local prediction models are set comprises:

Setting is based on the local prediction model of 30% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 50% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 75% boiler maximum continuous rating load section;

Setting is based on the local prediction model of 100% boiler maximum continuous rating load section.

4. the supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL according to claim 3 is characterized in that:

, said load parameter is adjusted during in the amplitude of variation of Preset Time internal loading parameter according to the local preset model under current greater than threshold value.

5. the supercritical unit control method for coordinating based on the multi-model PREDICTIVE CONTROL according to claim 4 is characterized in that:

Surpass preset crossing the border during time span when getting into another operating mode, carry out the switching scheduling between each local prediction model.

Images