JPH03223417A - Successive estimation of parameter in combustion control of continuous heating furnace and its utilization - Google Patents

Abstract

PURPOSE:To improve the accuracy of combustion control and to allow a stabler and more efficient operation by successively estimating the unknown parameters included in the simulation model of a continuous type heating furnace operation in accordance with the observed values of a very short term near the present time and a long term. CONSTITUTION:The combustion control of the continuous type heating furnace is executed by simulating the furnace operations at the point of the present time and from now-on by using a simulation model. The optimum operation quantity of the operations and the intended schedules, etc., of materials are determined in this way. The values of the unknown parameters contained in this model are successively estimated from the various observed values. Further, a short-term estimated values based on the short-term observation near the present time and a long-term estimated values based on the long-term observation are successively determined in parallel in this estimation. The weighted average values of the short- and long-term estimated values or the values obtd. after prescribed correction values are added to these values are used at the time of utilizing these estimated values. In addition, the values are so changed according to time intervals that the value is near to the short-term estimated value near the point of the present time and is more to more near to the long-term estimated value furtherer from the point of the present time. The combustion control of high accuracy can thus be executed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to combustion control of a continuous heating furnace.

[Conventional technology]

In the combustion control of a continuous heating furnace, the time series of fuel flow rate or temperature inside the furnace, the charging of the material to be heated into the furnace, the progress inside the furnace,
A simulation model (
A method of determining the operating amount by simulating the future operation of the furnace using a mathematical model (mathematical model formula) has been proposed in the past. The mathematical model used here generally includes parameters that vary randomly depending on the characteristics of the furnace, operating conditions, or surrounding environment.

Also, unknown disturbances that cannot be predicted by models cannot be ignored in actual operations. Calculation accuracy improves when the values of such parameters and disturbances are sequentially estimated using actual observed values such as furnace temperature and fuel flow rate as unknown parameters.

For example, in Japanese Patent Application No. 63-309573, the mathematical model for operation simulation is called a heating furnace simulator, and the unknown parameter there is the coefficient of heat loss in the heat balance equation for each zone in the furnace length. The heat loss is expressed by the following formula.

QLj=λA1・TFi+λ81...(1) Here, l: zone number Q Li: heat loss in the first zone TFi: furnace temperature in the first zone λAl, λ81: unknown parameter (λBi also includes unknown disturbance) ) Unknown parameter λA1. The value of λBi is sequentially estimated using observed values such as the furnace temperature and fuel flow rate each time the observed values are obtained at a predetermined period. Estimation methods include, for example, a weighted sequential least squares method in which the weight of observed values from the past to the present used for estimation calculations is reduced as the value is observed in the past, or a known method such as a fixed gain method. Use methods.

[Problem to be solved by the invention]

In the simulation of future operations as mentioned above,
In order to find the optimal operating amount, it is necessary to calculate several tens of times from the present moment to the future point in time when the material that is scheduled to enter the combustion zone on the furthest charging side (the zone where fuel is burned) is extracted from the furnace in the near future. It is necessary to perform a simulation for more than 1 minute.

On the other hand, in such a considerable time interval (several tens of minutes or more), the value of the unknown parameter of the mathematical model of the heating furnace is
It fluctuates considerably due to short-term fluctuations in operating conditions and material conditions, and in addition to these short-term fluctuations, long-term fluctuations also occur due to long-term fluctuations in these conditions and aging of equipment. For example, the value of the heat loss coefficient λB2 in the above-mentioned Japanese Patent Application No. 63-309573 during actual operation varies as shown in FIG.

Therefore, if the future point in the simulation is close to the present time, the influence of the short-term fluctuations mentioned above will be largely reflected,
After that, as the difference between the future point in time and the present time increases, the effect of short-term fluctuations will be reduced, and if this difference exceeds the range of influence of short-term fluctuations, only the effects of long-term fluctuations will be reflected. be.

However, in all conventional methods, there is only one type of parameter estimate. This estimate becomes more sensitive to the effects of short-term fluctuations as the correction gain of the estimate increases (in the case of the weighted sequential least squares method, this corresponds to decreasing the 1 weight more rapidly as you go back into the past). It is suitable for simulating a future point in time close to the present time, but is not suitable for simulating a future point in time far from the ffi point; on the other hand, if the correction gain is made small, the opposite is true. Therefore, when the correction gain is large, the simulation results at large time intervals will fluctuate unnecessarily due to the influence of short-term fluctuations as the real time changes, leading to frequent fluctuations in the manipulated variables and scheduled schedules, resulting in stable operation. If the correction gain is small, the effects of short-term fluctuations will not be reflected in simulations near the current moment, resulting in a decrease in control accuracy, which will be detrimental to energy savings. In other words, in a simulation over a large time interval from the present moment to a future point several tens of minutes away, parameter values suitable for each different point in time are determined using only one type of estimated value, as in the conventional method. In this case, any of the above problems will occur.

[Means to solve the problem]

The present invention solves the problems of the conventional method as described above, and uses short-term estimated values that change rapidly based on the most recent short-term observed values and long-term observed values as estimated values of unknown parameters. Two types of estimated values are sequentially obtained in parallel, a long-term estimated value that changes slowly based on

A weighted average value of the long-term estimated value or a value obtained by adding a predetermined correction value to it is used. The closer this value is to the present moment, the closer it is to the short-term estimated value, and the farther away from the present moment, the longer the long-term estimated value or the corrected value to it. So that the value is close to the added value,
Varies depending on the time interval from the current moment.

[Effect]

The present invention will be explained in detail below.

The subject of the present invention is a case where unknown parameters are sequentially estimated for conventional operation simulation in combustion control of a continuous heating furnace.

An example of the functional configuration of such combustion control is disclosed in Japanese Patent Application No. 63-3.
No. 09573. Its configuration is shown in FIG. The following will explain based on this figure.

In Fig. 5, the first actual performance processing function (first function) is started at a predetermined period and inputs plant actual results φ (heating furnace furnace temperature, fuel flow rate, etc.) to calculate material temperatures that cannot be directly observed. The estimated values are added to φ and set as φ', and the unknown parameters of the plant model are successively estimated to obtain the estimated values.

Next, the fuel flow setting calculation/high force function (second function) of 2 is as follows.
Started from the first function, the fuel flow rate setting value F for each zone of the heating furnace
Calculate sv and output the result.

On the other hand, the third target temperature increase curve optimization function (third function) successively optimizes the target temperature increase curve A* of the material used in the second function using a predetermined evaluation function.

In the second and third functions, a plant model is used to perform simulation calculations for predicting future furnace operations based on the current furnace status.

Here, unknown parameter sequential estimates are used.

The sequential estimation performed in the first function will be described below.

For sequential estimation, the plant model is generally expressed by the following equation that is linear with respect to the unknown parameter(s).

y (t) = A (t) x (t)
...(2) However, t: Time (sampling period unit) n: Number of unknown parameters, vector, transformed value of matrix y (Lx), x (t): Observed value at time t, or based on observed value Calculated value (y t i, x t Rrlx't)A
(t): Value of unknown parameter at time t (At good T
For example, in equation (1), n+ y+ Xg A is 2 w QL 1+ [TFl, 1 ]"
p [λAi, λBi]”.

In the present invention, as the estimated value of A, two types of estimated values are used in parallel: a short-term estimated value As based on the most recent short-term observation value, and a long-term estimated value AL based on the observed value over a long period of time.
It is calculated sequentially using the following formula.

As(t)=As(t-1)+ks(t) (y(t
,)-As(t-1)"x(t)) --
・(3) AL (t,)=AL (j-1)+kL
(t) (y(t) AL (t-1) x
(t)) ... (4) t= TaL, k 5(t)
, k L (t) (t Rnx')L;! Modification 'F
'I'J vector, using a known method such as the weighted sequential least squares method or the fixed gain method,
Calculate the value.

If we use a simple fixed gain method hereafter, ks(t)
= Psx(t)/(1+x(t)T Psx(
t)) ...The value of ks is determined as in (5). Here, Ps (εR) is a fixed diagonal matrix (gain matrix), and diagonal element psi (i=1.-, n; psi>O
) determines the value of the modified gain of each element of As. The same is true for kL, and in equation (5), kL is used instead of ks and Ps.
.

It is sufficient to set it as PL(εgoodnxn).

Now, here, the value of the matrix Ps is set to a size that allows the short-term estimated value As to follow short-term fluctuations of the unknown parameter A sufficiently quickly, while the value of the matrix PL is set so that the long-term estimated value AL
is set sufficiently small so that A is not affected by short-term fluctuations in A and slowly follows only long-term fluctuations.

Next, in the simulation of future operation at actual time t, the value of the estimated value A(t, u) of A used for calculation at future time (10 u) on the simulation is determined by the following equation.

A(t,,u)=[(u&lx-um)As(t)
+ur (AL (t)+h71) ] 8m...(6) However, u so = +5 in (u wax, u ) Here, uma is the upper limit of U error, and is about half of the approximate fluctuation period of As. shall be. Also, hACt Rnx”) is the AL correction value 6.hAt
7) Although the value may be 0, the value A(τ) of the unknown parameter at the actual time τ in the future from t becomes the long-term estimated value AL(t,
), for example, As
Set a predetermined value approximately equal to the amplitude of short-term fluctuations. For example, in equation (1), if the value of λBi increases for some reason in the future, the heat loss will increase and the temperature of the material will not rise as per the previous simulation. To prevent undercooking that may occur in such cases, λB
It is preferable to set an appropriate positive number as a correction value for 1.

As can be seen from the above equation, the estimated value A (t, u) used in the simulation at the future point in time (t + u) is the short-term estimated value As(t) and the long-term estimated value + correction value (AL (t
)+hA), the smaller U is, the closer it is to the former, and the larger U is, the closer it is to the latter.

A graph of A(t, u) versus U is shown in FIG.

〔Example〕

Hereinafter, as an example of the present invention, Japanese Patent Application No. 63-30957
The results of a control simulation conducted in accordance with the heating furnace combustion control method described in No. 3 will be described. The target plant here is a continuous heating furnace for steel billet slabs in a continuous hot rolling mill of a steel works. 2
There are three heating zones (2Hz). The simulation was performed based on the same actual operation data as in FIG. 1, and the value of the heat loss coefficient (unknown parameter) shown in equation (1) was changed as in the actual operation. The control method here is an optimal control method by minimizing a predetermined evaluation function, and this evaluation function is
The average temperature of each extraction slab is above the target temperature.

And the uneven heat inside the slab at that time (temperature difference between the surface and center)
The fuel flow rate was set to be minimized under the constraint that the temperature was within a predetermined value (50° C.).

3a, 3b, 3c, and 3d show the results when the present invention is incorporated in the control described in 2.
Figures 4a, 4b, 4c, and 4d show the results of a conventional example in which the long-term estimated value was made exactly the same as the short-term estimated value without incorporating the short-term estimated value, and was frequently changed. In this plant,
In order to minimize the fuel flow rate, within the range that does not exceed the upper limit allowed by furnace temperature equipment and slab quality, 3) 1z
It is effective to increase the amount of fuel for 2Hz as much as possible and decrease the amount of fuel for 2Hz as much as possible. Example incorporating the present invention (third example)
In Figures a to 3d), such control is almost achieved while observing the above constraints.

However, in the conventional example (FIGS. 4a to 4d), compared to the case where the present invention is implemented, the fuel flow rate at 3 Hz is small and the fluctuation is large, but the fuel flow rate at 2 Hz is greatly increased and the thermal efficiency is low. Therefore, the fuel flow rate of the entire furnace increased by about 4% in the examples shown in FIGS. 4a to 4d compared to the examples shown in FIGS. 3a to 3d. This demonstrated that the present invention is effective in stable operation and energy saving in heating furnace combustion control.

〔Effect of the invention〕

As described above, according to the present invention, in the combustion control of a continuous heating furnace, the influence of the most recent observed values is greatly reflected in the simulation at a point in the near future to improve control accuracy, while In future simulations, the influence of long-term observed values is reflected to prevent unnecessary fluctuations in simulation results due to fluctuations in short-term estimated values, and by adding correction values, future fluctuations in unknown parameters can be reduced. Prevent serious problems (such as under-baking of materials) that can result from mispredictions caused by This makes it possible to save energy, stabilize operations, and prevent under-cooking in combustion control.

[Brief explanation of drawings]

Fig. 1 is a graph showing the fluctuation of unknown parameters, Fig. 2 is a graph showing the correlation between U and A(t, u) in the method of the present invention, Figs. 3a, 3b, 3c, and 3d.
Figure 1 shows graphs representing simulation results when the present invention is used, Figures 4a, 4b, 4c, and 4.
FIG. 1 is a graph showing the simulation results when the present invention is not used, and FIG. 5 is a block diagram showing the functional configuration of smoking control. Voice figure 300.0 400.0 500.0 600.0 Mm (mtn) 700.0 Voice 38 figure Voice 3b figure Saitaku (min) 4d figure Voice 4b figure Sight review (min) Voice 3c figure Time opening (min ) Voice 4c illustration Ginseki (min)

Claims (1)

[Claims] In combustion control of a continuous heating furnace, the time series of fuel flow rate or furnace temperature, charging of the material to be heated into the furnace, progress in the furnace,
a simulation model for simulating the operation of a furnace given a scheduled schedule for extraction from the furnace; By performing simulations, we can determine the optimal manipulated variables for fuel flow rate or furnace temperature, material schedule, etc. At the same time, the values of unknown parameters included in the model can be calculated sequentially from observed values such as furnace temperature and fuel flow rate. In the estimation method: Regarding unknown parameters, two types of estimated values are successively obtained in parallel: a short-term estimated value based on the most recent short-term observed values, and a long-term estimated value based on long-term observed values, and the two types of estimated values are sequentially obtained during the simulation. For the parameter values at each future point in time, we use the weighted average of the short-term estimated value and the long-term estimated value, or the value obtained by adding a predetermined correction value to it. Sequential parameter estimation and its use in combustion control of a continuous heating furnace, characterized in that the value is changed depending on the time interval from the current time so that the value becomes closer to the long-term corrected estimated value as the distance from the current time increases. Method.