CN100472373C - Real time operation optimizing method for multiple input and multiple output continuous producing process - Google Patents

Abstract

The method of optimizing operation in real time is related to continuous process with multiple inputs and outputs. The method considers multiple key operation conditions in production procedure as variables to be optimized, and considers technical indexes associated to key operation conditions as objective functions. Based on historical data of the key operation conditions and technical indexes, using correlation integral method or other method calculates out gradient vector between the conditions and technical indexes in current time online. Then, based on the gradient vector, the method determines direction of adjusting operation conditions. When the gradient vector is a positive or negative value, the method adjusts key operation conditions to make gradient vector change along direction towards zero, to let technical indexes reach optimization. The invention simplifies optimizing control procedure, providing features of simple operation, high interference immunity etc.

Description

A kind of real time operation optimizing method of multiple input and multiple output continuous producing process

Technical field

The invention belongs to the control system technical field, relate to a kind of multiple-input and multiple-output (MultipleInput/Multiple Output, MIMO) real time operation optimizing method of continuous flow procedure.

Background technology

To a continuous feed, the continuous production run of continuous product output, can represent with Fig. 1: A is upstream processing unit (plant) group among the figure, and its product is the raw material of processing unit (plant) B, and the product of processing unit (plant) B is the raw material of downstream unit C.Why A is referred to as the device group, is because may a more than cover upstream device be that B supplies raw materials.Because the product of B may be a kind of incessantly, its product also may offer a plurality of downstream units as raw material, thus its downstream device be referred to as device group C.Be worth pointing out that it is continuous that the serial connection of this device material flows.For processing unit (plant) B, have in some productions and can adjust crucial factor, just controlled key operation condition U, and some technical indicator J of these operating conditionss and processing unit (plant) B are as energy consumption, economic benefit, product yield is relevant, and U can regard the input of production run as here, can have one or more, J sees the output of production run as, also can have one or more.The task of operation optimization is how to adjust these operating conditionss aborning, and (maximum is as economic benefit to make the technical indicator (can be, or a plurality of combinations) of appointment in section sometime reach optimum; Or minimum, as energy consumption), as shown in Figure 2.Be worth pointing out that objective function may switch according to the needs of producing.To the operation optimization of simple target functional procedure, generally with following formal description:

$J^{*}=J\left(U^{*}\right)=J\left(U\right)_U^{\max }---(1-1)$

In the formula, J is an objective function, J ^*Be the objective function optimal value, U is an operating conditions, or claims the tuning variable.U ^*Optimum value for the tuning variable.

Because the funtcional relationship J (U) of U and J is a unknown function usually, in order to obtain U ^*, two kinds of methods commonly used are arranged: mathematical model and real-time online search procedure.

1. modeling

The basic thought of modeling is the mathematical model of setting up in advance between objective function J and tuning variable U, then according to this model and constraint condition, obtains U with non-linear or linear programming ^*

In setting up the process of mathematical model,, can be divided into modelling by mechanism and experience modeling two classes according to the difference of modeling principle.

So-called modelling by mechanism is exactly the running mechanism equation each several part equipment in the total system, by flowage structure, utilizes material balance, principle of energy balance to combine, thereby constructs the math equation that a cover meets concrete production run.Determine relation between objective function and tuning variable U by the inputoutput of system, price etc. at last.

But it is too complicated to work as the process that is described, and perhaps mechanism itself is unclear, and perhaps fundamental equation is inaccurate, often makes modelling by mechanism work be difficult to carry out.In addition, the mechanism model of a system does not generally have ubiquity, even changes when converted products, or flow process changes slightly, just must make modification or makes a new start model.

The experience modeling relies on a large amount of tests or day-to-day operation report data to be basis, the tuning variable of tectonic system and the empirical relationship between objective function exactly.The advantage of this modeling is: simple, have a ubiquity.No matter how complicated or different process or system be, all available same simple means modeling, and do not need special process knowledge and priori equation.

But the reliability of this method is relatively poor.When model is online when using condition of work departs from or exceed the operating mode that modeling data gathers, model will produce very big error and can't practicality, and the small change of process equipment may cause the very big variation of model structure and makes modeling work all that has been achieved is spoiled.

2. on-line search method

Search procedure is a kind of method of ubiquity, and is irrelevant with detailed process.Its basic thought is the numerical value that changes the tuning variable online, the situation of change of object observing function, thus determine whether the tuning change direction is correct.Say in principle, many nonlinear programming approach, as Fibonacci method, method of steepest descent etc. can both onlinely be used.But these methods are often very strong to the susceptibility of disturbing.As everyone knows, under normal conditions, objective function is not only the function of tuning variable, also is the function of other uncontrolled variables (environmental variance).Thereby when objective function changed, we be difficult to judge that it is to result from the variation of tuning variable, the still effect of disturbing.In present on-line search method, general all tuning variable and objective function as unique cause-effect relationship, thereby when having external disturbance, might do the judgement that makes mistake, even make action reversed.

No matter be worth pointing out, be modelling or direct search method, generally is to be based upon on the mathematical description basis of 1-1.And the relation between reflection objective function J and the U is defined as algebraic relation, does not also comprise the external interference item. thereby the algorithm of Dao Chuing only is applicable to static, glitch-free system in principle thus.

In real process, situation is more complex.At first, the tuning variable not only logically has cause-effect relationship with objective function, and also exists dynamic process in time, that is to say when the tuning variable changes, and objective function does not change at once, but a transient process is arranged.Secondly, from the actual conditions of many industrial processs, objective function usually is among the pulsation of dipping and heaving, is difficult to find static situation, and this is normally owing to those can not be surveyed, uncontrollable strong disturbance causes.For example in the production run, the variation of raw material composition is uncontrollable often, because the difficulty that online composition detects, these variablees also are immesurable usually.And on the other hand, many processes are but very responsive for the variation of raw material composition, and the variation of composition as a result usually will exceed controlled temperature, the pressure factors effect to objective function for the influence of objective function, can reach tens of times more than sometimes.So, the caused objective function variation of tuning variable change composition often will " flood " among the interference of raw material composition to objective function.Under the situation of this dynamic disturbances, it is powerless that traditional method just seems.

Summary of the invention

Technical matters to be solved by this invention be big at the existing process complexity of the real time operation optimizing of multiple input and multiple output continuous producing process in the prior art, difficulty, be subjected to that disturbing factor is many, adaptive surface is narrow, the defective of weak effect, a kind of real time operation optimizing method of multiple input and multiple output continuous producing process is provided, and it is wide, effective to simplify the optimal control process of production run, simple to operate, strong interference immunity, highly versatile, adaptation.

The technical scheme that solution the technology of the present invention problem is adopted is to adopt a kind of real time operation optimizing method of multiple input and multiple output continuous producing process, this method is an optimization variable with a plurality of key operation conditions in the production run, with the one or more technical indicators that are associated with the key operation condition is objective function, dynamic history data according to production run key operation condition and technical indicator, online calculate between key operation condition and the technical indicator gradient vector in the current time, determine the adjustment direction of operating conditions then according to this gradient vector, when gradient vector be negative value or on the occasion of the time, all will adjust the key operation condition makes gradient vector change to zero direction, make technical indicator reach optimum, the calculating of this gradient is online ongoing, no matter whether technical indicator reaches optimum, in case find that gradient is non-vanishing, then adjust the key operation condition gradient vector is changed to zero direction, to realize the tracking of optimum point.It is exactly the process of key operation condition optimizing to the process that zero direction changes that adjustment key operation condition makes gradient vector, because the optimum point of production run may change in time, therefore this optimizing process is online ongoing.

The method that online calculates between key operation condition and the technical indicator in the gradient vector of current time is to utilize correlation integral technology or other possible methods (as the dynamic model identifying approach) that the dynamic history data of production run key operation condition and technical indicator are carried out computing.

The foundation of correlation integral real-time optimization method is the correlation integral theory.Correlation integral is a kind of computing relevant with stochastic process.In the correlation integral theory, objective function, interference, optimization variable are regarded as stochastic process, and optimization variable is that average is controlled, usually determine objective function earlier

This objective function should be online computable or online measurable, and objective function can be expressed as so:

$\tilde{J}\left(t\right)=f(\tilde{u}\left(t\right),\tilde{p}\left(t\right),t)$

Wherein, Be m dimension average optimized controllable variable

$\tilde{u}\left(t\right)=\left[\begin{array}{c}{\tilde{u}}_1\left(t\right)\\ {\tilde{u}}_2\left(t\right)\\ \·\\ \·\\ \·\\ {\tilde{u}}_m\left(t\right)\end{array}\right]$

For disturbing, f is a unknown mappings; The optimal objective function definition is:

$\tilde{J}*\left(t\right)=\underset{E\left\{\tilde{u}\left(t\right)\right\}}{\max }E\left\{f[E\left\{\tilde{u}\left(t\right)\right\},\tilde{p}\left(t\right),t]\right\}$

Here

Be the average of tuning variable, it can be the setting value of basic unit's controller or valve position etc.Multivariable hereto optimization problem can prove that the objective function average is to the gradient of tuning variable average under certain condition

Satisfy following formula:

$k_{\mathrm{uJ}}=k_{\mathrm{uu}}\frac{\mathrm{dEf}}{\mathrm{dE}\left\{\tilde{u}\left(t\right)\right\}}+\mathrm{\ϵ}\left(t\right)$

In the formula, ε (t) is that average is zero noise item, and k _UJBe cross-correlation integral vector between tuning variable and objective function, be defined as

$k_{\mathrm{uJ}}=\left[\begin{array}{c}{k}_{u1J}\\ k_{u2J}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}}\end{array}\right],$ $k_{\mathrm{uiJ}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})J\left(t\right)\mathrm{dtd\τ},i=\mathrm{1,2},\·\·\·,m$

k _UuFor tuning variable autocorrelation integral matrix, be defined as

$k_{\mathrm{uu}}=\left[\begin{array}{cccc}{k}_{u1u1}& k_{u1u2}& \·\·\·& k_{u1\mathrm{um}}\\ k_{u2u1}& k_{u2u2}& \·\·\·& k_{u2\mathrm{um}}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{\mathrm{umu}1}& k_{\mathrm{umu}2}& \·\·\·& k_{\mathrm{umum}}\end{array}\right],$ $k_{\mathrm{uiuj}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})u_j\left(t\right)\mathrm{dtd\τ},j=\mathrm{1,2},\·\·\·,m$

According to formula

$k_{\mathrm{uu}}=\left[\begin{array}{cccc}{k}_{u1u1}& k_{u1u2}& \·\·\·& k_{u1\mathrm{um}}\\ k_{u2u1}& k_{u2u2}& \·\·\·& k_{u2\mathrm{um}}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{\mathrm{umu}1}& k_{\mathrm{umu}2}& \·\·\·& k_{\mathrm{umum}}\end{array}\right]$

$k_{\mathrm{uiuj}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})u_j\left(t\right)\mathrm{dtd\τ}$ Real-time measurement values u with optimization variable _i(t) (i=1,2 ..., m) the autocorrelation integral matrix k of calculation optimization variable _UuT wherein, M is the integration constant greater than 0;

According to formula

$k_{\mathrm{uJ}}=\left[\begin{array}{c}{k}_{u1J}\\ k_{u2J}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}}\end{array}\right]$

$k_{\mathrm{uiJ}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})J\left(t\right)\mathrm{dtd\τ},(i=\mathrm{1,2},\·\·\·,m)$ With real-time measurement values J (t) the calculation optimization variable of objective function and the cross-correlation integral vector of target variable; T wherein, M is the integration constant greater than 0;

In the above-mentioned formula, u _i(t), (i=1,2 ..., m), J (t) is respectively the measured value of tuning variable and objective function.As seen k _UJ, k _UuCan calculate by the observed reading of tuning variable and objective function, so according to formula $k_{\mathrm{uJ}}=k_{\mathrm{uu}}\frac{\mathrm{dEf}}{\mathrm{dE}\left\{\tilde{u}\left(t\right)\right\}}+\mathrm{\ϵ}\left(t\right)$ Obtain the gradient of objective function

(the gradient of available least square method estimation objective function

Calculate after the gradient of objective function, can be with the new settings value u of direct iterative computation tuning variable _s(l+1)

According to formula $u_s(l+1)=u_s\left(l\right)+\mathrm{\α}\frac{\mathrm{dEf}}{\mathrm{dE}\left\{\tilde{u}\left(l\right)\right\}}$ The new settings value u of calculation optimization variable _s(l+1); In the formula, α is a constant, if optimization aim is a maximizing, then is taken as the number greater than 0, if optimization aim is a minimal value, is taken as the number less than 0.

Online the continuing of this iterative process carries out, and is zero until gradient.

For multiobject situation, similar conclusion is arranged also.

By last, it is as follows to draw concrete further technical scheme of the present invention:

1. according to the needs of optimised process, determine that a plurality of technical indicators that will optimize are objective function J ₁, J ₂..., J _n, these objective functions must onlinely can calculate or measure.Construct a virtual integrated objective function J=σ ₁J ₁+ σ ₂J ₂+ ... ,+σ _nJ _n, σ here ₁, σ ₂..., σ _nFor the weighted number of each objective function, according to technological requirement, value is between 0 to 1, it may be noted that therefore weighted value may be time dependent owing to objective function in producing may switch here.

According to the requirement of production technology, determine the key operation condition u that will optimize ₁, u ₂..., u _mAs the variable that will optimize.

The key operation condition that will optimize is carried out conventional definite value control, and its setting value adopts the correlation integral technology to calculate by the optimal control computing machine, according to the concrete condition and the requirement of technological process, carries out the adjustment of a setting value every cycle regular hour;

It is to be that DCS or conventional instrument carry out conventional definite value control to the operating conditions that will optimize by the distributing system computing machine earlier that the key operation condition that will optimize is carried out conventional definite value control, and its setting value adopts the correlation integral technology to calculate by the optimal control computing machine, carries out the adjustment of a setting value every cycle regular hour (time cycle is determined by the speed of concrete technological process).

2. gather the data of key operation condition and each technical indicator (being each objective function).Its method is according to concrete process time characteristic, foundation has certain data window width, and (time width of this database should be greater than this process optimization variable to more than 3 times of objective function settling time, be generally 8-18 hours) the real-time data acquisition data system, usually this system is made of Distributed Control System (DCS) (being DCS), to obtain the historical data of production run key operation condition and technical indicator.The key operation condition is gathered and each technical indicator is the data of each objective function every certain sampling time interval (speed of this time interval by technological process decides, be generally 30--90 second) by this system.Data storage in the data window is whenever adopted sample one time among database, data window just moves forward a sampling time, that is to say, the oldest data are abandoned, and up-to-date data are added among the database, are illustrated in figure 3 as the example of 2 operating conditionss.

3. after data sampling finishes, each operating conditions is carried out autocorrelation integral matrix k _UUCalculating: be provided with m operating conditions

$k_{\mathrm{UU}}=\left\{\begin{array}{cccc}{k}_{11}& k_{12}& \·\·\·& k_{1m}\\ k_{21}& k_{22}& \·\·\·& k_{2m}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{m1}& k_{m2}& \·\·\·& k_{\mathrm{mm}}\end{array}\right\}$

In the formula: $k_{\mathrm{ij}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)u_j(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

I, j=1,2 ... m T, M are integration constant.

4. calculate the cross-correlation integral matrix K between each performance variable and the technical indicator _UJ: be provided with n objective function

$K_{\mathrm{UJ}}=\left[\begin{array}{c}{K}_{\mathrm{UJ}1}\\ K_{\mathrm{UJ}2}\\ \·\\ \·\\ \·\\ K_{\mathrm{UJn}}\end{array}\right]$

In the formula:

$K_{\mathrm{UJ}1}=\left[\begin{array}{c}{k}_{u1J1}\\ k_{u2J1}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}1}\end{array}\right]......K_{\mathrm{UJn}}=\left[\begin{array}{c}{k}_{u1\mathrm{Jn}}\\ k_{u2\mathrm{Jn}}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJn}}\end{array}\right]$

$k_{\mathrm{uiJs}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)J_s(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

(i＝1，2，…，m；s＝1，2，…，n)

T, M are integration constant.

According to the correlation integral theory, above M should be greater than the time constant from optimization variable to the objective function maximum, and T is taken as 1-5 M doubly.

5. according to the autocorrelation integral matrix of above operating conditions and the cross-correlation integral vector between performance variable and each technical indicator, calculate operating conditions to the gradient vector between comprehensive technical indexes Earlier obtain K by calculating following linear equation ^d:

K _UJ＝K _UUK ^d

And

$K^d=\left[\begin{array}{c}{K}_{J1}^d\\ K_{J2}^d\\ \·\\ \·\\ \·\\ K_{\mathrm{Jn}}^d\end{array}\right]$

Integrated objective function J is to the gradient of operating conditions

For

$K_{\mathrm{\σ}}^d={\mathrm{\σ}}_1{K}_{J1}^d+{\mathrm{\σ}}_2{K}_{J2}^d,\·\·\·,K_{\mathrm{Jn}}^d;$

6. according to the gradient vector of resulting operating conditions to comprehensive technical indexes

Calculate the change direction of operating conditions, its principle is that then this operating conditions is at present at optimum state if the gradient of calculating is zero; If non-vanishing, carry out the operating conditions adjustment according to the size and the direction of gradient.For example: present operating conditions value is known, obtains adjusted operating conditions value according to the methods below, requires the comprehensive technical indexes maximization:

In the formula:

$\left[\begin{array}{c}{u}_1\left(l\right)\\ u_2\left(l\right)\\ \·\\ \·\\ \·\\ u_m\left(l\right)\end{array}\right]$ Be the original m operating conditions value in (l step), and $\left[\begin{array}{c}{u}_1(l+1)\\ u_2(l+1)\\ \·\\ \·\\ \·\\ u_m(l+1)\end{array}\right]$ Be the adjusted new operating conditions value in (l+1 step).

$\left[\begin{array}{c}{k}_{1\mathrm{\σ}}^d\\ k_{2\mathrm{\σ}}^d\\ \·\\ \·\\ \·\\ k_{\mathrm{m\σ}}^d\end{array}\right]=K_{\mathrm{\σ}}^d$

α ₁, α ₂..., α _mBe m positive constant (if asking the technical indicator maximal value).We are example with the production run of an operating conditions in the hope of the technical indicator maximal value, the method that diagram is adjusted.See Fig. 4, Fig. 5, shown in Figure 6.Illustrated among Fig. 4 when calculating gradient with correlation-integral method and be zero, technical indicator is a maximal value, and operating conditions need not be adjusted this situation.Illustrated among Fig. 5 when calculating gradient with correlation-integral method when negative, operating conditions should reduce, and index just can develop skill.Having illustrated when calculating gradient with correlation-integral method among Fig. 6 is timing, and operating conditions should increase, and index just can develop skill.

Obviously, to i operating conditions u _i, adjust step-length at every turn and be

As long as get suitable α _iValue, if promptly ask the maximal value of technical indicator, get α _iFor on the occasion of, otherwise get negative value, just can adjust the size and Orientation of step-length.

After adjustment was finished, at regular intervals (30-90 seconds) carried out data sampling once more, return step 3.

Step 3 to 6 process is online ongoing, can make each operating point finally reach optimum point.Even if it is to be noted that each gradient has arrived zero, but step 2 to 5 also should constantly be carried out, this is because the funtcional relationship between technical indicator and the operating conditions may change along with the time (as the variation of feedstock property, the transformation of device, Deng), need observe constantly whether gradient is zero,, adjust at any time if change.As shown in Figure 7: change when the relation of technical indicator and operating conditions (changes as raw material) for a certain reason, make when current operating point no longer is optimum point, can find that gradient is non-vanishing, according to the method for correlation integral that operating conditions is approaching to optimum point.

This just makes this method can find at any time whether production run has departed from optimum point, and carries out the tracking of optimum point.

Correlation integral real-time optimization method has the following difference characteristics of optimisation technique in the past:

● the present invention has adopted correlation-integral method to carry out gradient calculation, it is that the fluctuation data of (being in the data window) in a period of time are calculated according to operating conditions and technical indicator, therefore different with model method, do not require that when Optimizing operation production run is in static state, can be in fluctuation status, and according to the correlation integral theory, the statistical nature of these fluctuations can be unknown.

● can not need to set up in advance the static state and the dynamic model of process, significantly reduce the complicacy of production run operation optimization.Traditional method is that modelling by mechanism or statistical modeling all are the funtcional relationships of managing in advance to obtain between technical indicator J and the operating conditions U, gets on to obtain optimal operation conditions on the basis of this funtcional relationship then.If system very complex just is difficult to obtain model accurately, required cost is also very high.As long as and the present invention notices near the information that current operating point is, need not set up large-scale model and promptly can be optimized operation system, can save complicated process model building.

● this law has very strong adaptive performance.That is to say,, make when operating conditions departs from optimum point that the present invention can find this departing from, and automatically operating conditions is readjusted on the optimum when production run changes for a certain reason.The performance of this automatic tracking optimum point has important practical and is worth in actual production.Because the relation in many production runes between operating conditions and the technical indicator is actually the variation along with feedstock property, equipment aging, substitute improvement of catalyzer etc. and changing.And traditional modeling just need be to production run modeling again, or model is revised.

● this method is a kind of method of versatility.As long as production run is continuous, the technical indicator that optimize can on-line measurement or calculating, can use this method, and is irrelevant with concrete production run, adaptable wide range.And traditional modeling must carry out modeling at a certain concrete device, and the Optimization Model of foundation has particular applicability.

● utilize the fluctuation naturally of the normal operation of process to carry out work, need not in addition process to be added test signal, thus very little to the interference of process operation.

● this method has very strong noiseproof feature, even in dynamic strong jamming, be the variation that causes technical indicator of other factors such as feedstock property greater than still can operate as normal under the mal-condition of useful signal (being caused the variation of objective function by optimised operating conditions), this strong anti-interference has great importance in actual applications.

Adopting the core of correlation integral real-time optimization method among the present invention is to utilize the gradient of the principle calculating target function of correlation integral technology to operating conditions, carry out the adjustment of operating conditions then according to gradient, and the adjustment of the calculating of gradient and operating conditions is constantly online carries out.Therefore, no matter use what computing method, if can online calculating operation condition and the gradient of objective function can carry out on-line optimization to device with this principle.Therefore adopt the gradient of same online calculating operation condition of energy of other possible methods such as dynamic model identifying approach and objective function, and then real-time optimization is carried out in the continuous production operation of device.

Description of drawings

The continuous production run block scheme of continuous feed of Fig. 1, continuous product output

Fig. 2 is the operation optimization schematic flow sheet of continuous flow procedure

Fig. 3 has the data window data processing of 2 operating conditionss and gathers synoptic diagram

Curve map when Fig. 4 is zero for the gradient of the curve of operating conditions and technical indicator

Fig. 5 is the curve map of gradient when negative of the curve of operating conditions and technical indicator

Fig. 6 is the curve map of timing for the gradient of the curve of operating conditions and technical indicator

Fig. 7 is the relation of operating conditions and the technical indicator curve map that the gradient of curve in period changes thereupon that changes

Fig. 8 is optimized the control chart of operation for the present invention adopts computing machine to continuous flow procedure

Fig. 9 is an ARGG device reaction regenerative system process flow diagram

Figure 10 is a ketone benzene de-oiling dewaxing integrated unit process chart

Among the figure: 1-riser reactor 2-cyclone separator 3-settler 4-catalyst regenerator 5-external warmer, 6-fractionating system

Embodiment

It below is non-limiting examples of the present invention, when implementing, require production run to adopt computing machine to be optimized control, by distributing system computing machine (being DCS) or conventional instrument the operating conditions that will optimize is carried out conventional definite value control earlier, and its setting value adopts method of the present invention to calculate by the optimal control computing machine, carry out the adjustment of a setting value every cycle regular hour, the speed of the cycle of this adjustment and concrete technological process decides.As shown in Figure 8.

The online correlation integral optimal control of embodiment 1:ARGG device

The ARGG device is that petrochemical plant is used for the oil plant of low value is cracked into liquid hydrocarbon, the gasoline of high value, the device of diesel oil, belongs to continuous flow procedure.ARGG device reaction regenerative system process flow diagram as shown in Figure 9.Feedstock oil and freshening gasoline, recycle stock and reprocessed oil slurry from the tank field are mixed, and atomizing sprays into riser reactor 1 in riser reactor 1 bottom, enters reactive system.

The reaction raw materials of atomizing and the steam that sprays into are mixed in riser reactor 1 bottom with the high temperature catalyst from catalyst regenerator 4, carry out catalytic cracking reaction along riser reactor 1 rising, reacted oil gas enters cyclone separator 2 rapidly with the catalyst separation of oil gas and solid granular together with catalyzer in the riser exit, isolated oil gas is sent into fractionating system 6.

Isolated catalyzer enters catalyst regenerator 4, and these catalyst surfaces have the carbon distribution that produces in course of reaction, and these carbon distributions burn in regenerator and remove, and this process is called burns.The superfluous heat that produces in the burning process is taken away by external warmer 5.Catalyzer behind coke burning regeneration enters riser reactor 1 bottom again, mixes with raw material again and reacts.In order to keep the catalyzer of activity of such catalysts and supplement consumed, the outside also replenishes raw catelyst to catalyst regenerator.

Oil gas from separation vessel enters fractionating system 6, the separation through this system, output liquid hydrocarbon, gasoline, light diesel fuel.And the reprocessed oil slurry of output and recycle stock and part gasoline turn back to riser reactor 1.

In the application example of this device, following key operation variable is elected optimization variable as:

● the riser reactor outlet temperature

● promote dry gas flow in advance

● feeding temperature

● the terminator flow

● the freshening steam oil ratio (SOR)

● the recycle stock flow

● the reprocessed oil slurry flow

● the raw catelyst feeding quantity

And optimised technical indicator (objective function) has 5:

● liquid hydrocarbon yield

● yield of gasoline

● diesel yield

● total liquid is received yield

● install total economic benefit

The structure of optimal control system as shown in Figure 8, the set-point control system is made up of the Distributed Control System (DCS) of HONEYWELL TPS3000, and optimal control is born by APP (application process processor) computing machine of HONEYWELL TPS3000.Requirement according to current technology is selected one as the current technical indicator that will optimize from above five technical indicators.And carry out according to following steps

1. name each key operation variable

u ₁=riser reactor outlet temperature

u ₂=promote dry gas flow in advance

u ₃=feeding temperature

u ₄=terminator flow

u ₅=freshening steam oil ratio (SOR)

u ₆=recycle stock flow

u ₇=reprocessed oil slurry flow

u ₈=raw catelyst feeding quantity

Name each objective function:

J ₁=liquid hydrocarbon yield

J ₂=yield of gasoline

J ₃=diesel yield

J ₄=total liquid is received yield

J ₅=install total economic benefit

2. utilize HONEYWELL TPS3000 distributing system and APP computing machine to set up a data acquisition system (DAS) based on OPC.Set up in the APP computing machine that to have the data window width be 8 hours real-time data base.Gathered the data of key operation condition and each technical indicator (each objective function) every 60 seconds.Data storage in the data window is whenever adopted sample one time among database, data window just moves forward a sampling time, that is to say, the oldest data are abandoned, and up-to-date data are added among the database.

3. after data sampling finishes, each operating conditions is carried out autocorrelation integral matrix k _UUCalculating: existing 8 operating conditionss

$k_{\mathrm{UU}}=\left\{\begin{array}{cccc}{k}_{11}& k_{12}& \·\·\·& k_{18}\\ k_{21}& k_{22}& \·\·\·& k_{28}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{81}& k_{82}& \·\·\·& k_{88}\end{array}\right\}$

In the formula: $k_{\mathrm{ij}}={\∫}_{-3600}^{3600}\frac{1}{2T}{\∫}_{-10800}^{10800}{u}_i\left(\mathrm{\λ}\right)u_j(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

Upper lower limit value 3600 in the above-mentioned integration ,-3600,10800, the-the 10800th was determined according to response time roughly from optimization variable to the objective function passage of this process.

4. calculate the cross-correlation integral matrix K between each performance variable and the technical indicator _UJ: existing 5 objective functions

$K_{\mathrm{UJ}}=\left[\begin{array}{c}{K}_{\mathrm{UJ}1}\\ K_{\mathrm{UJ}2}\\ \·\\ \·\\ \·\\ K_{\mathrm{UJ}5}\end{array}\right]$

In the formula:

$K_{\mathrm{UJ}1}=\left[\begin{array}{c}{k}_{u1J1}\\ k_{u2J1}\\ \·\\ \·\\ \·\\ k_{u8J1}\end{array}\right]......K_{\mathrm{UJ}5}=\left[\begin{array}{c}{k}_{u1J5}\\ k_{u2J5}\\ \·\\ \·\\ \·\\ k_{u8J5}\end{array}\right]$

$k_{\mathrm{uiJs}}={\∫}_{-3600}^{3600}\frac{1}{2T}{\∫}_{-10800}^{10800}{u}_i\left(\mathrm{\λ}\right)J_s(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

(i＝1，2，…，8；s＝1，2，…，5)

5. according to the autocorrelation integral matrix of above operating conditions and the cross-correlation integral vector between performance variable and each technical indicator, calculate operating conditions to the gradient vector between comprehensive technical indexes

Earlier obtain K by calculating following linear equation ^d:

K _UJ＝K _UUK ^d

And

$K^d=\left[\begin{array}{c}{K}_{J1}^d\\ K_{J2}^d\\ \·\\ \·\\ \·\\ K_{\mathrm{Jn}}^d\end{array}\right]$

Integrated objective function J is to the gradient of operating conditions

For

$K_{\mathrm{\σ}}^d={\mathrm{\σ}}_1{K}_{J1}^d+{\mathrm{\σ}}_2{K}_{J2}^d,\·\·\·,{{\mathrm{\σ}}_{5}K}_{J5}^d$

In the formula, σ ₁, σ ₂..., σ ₅Be 50 or 1 constant, decide according to the objective function of current optimization.For example, current requirement optimization be liquid hydrocarbon yield, then get σ ₁=1, all the other are taken as 0, other situations and the like.

6. according to the gradient vector of resulting operating conditions to comprehensive technical indexes

Calculate the change direction of operating conditions, its principle is that then this operating conditions is at present at optimum state if the gradient of calculating is zero; If non-vanishing, carry out the operating conditions adjustment according to the size and the direction of gradient.For example: present operating conditions value is known, obtains adjusted operating conditions value according to the methods below, requires the technical indicator maximization:

In the formula:

$\left[\begin{array}{c}{u}_1\left(l\right)\\ u_2\left(l\right)\\ \·\\ \·\\ \·\\ u_8\left(l\right)\end{array}\right]$ Be 8 original operating conditions values, and $\left[\begin{array}{c}{u}_1(l+1)\\ u_2(l+1)\\ \·\\ \·\\ \·\\ u_8(l+1)\end{array}\right]$ Be adjusted new operating conditions. $\left[\begin{array}{c}{k}_{1\mathrm{\σ}}^d\\ k_{2\mathrm{\σ}}^d\\ \·\\ \·\\ \·\\ k_{8\mathrm{\σ}}^d\end{array}\right]=K_{\mathrm{\σ}}^d$

α ₁, α ₂..., α ₈Be 8 positive constants, the size of these constants is relevant with the speed of convergence of optimal control, need adjust at the scene.Adjustment result according to the scene all gets 0.001.

7. got back to for the 3rd step.

According to the test of real system, liquid towards hydrocarbon yield, total liquid receive yield and these three kinds of technical indicators of device economic benefit obtain following result:

The contrast of the technical indicator before and after optimizing

The optimisation technique index	Before the optimization	After the optimization	Optimize the back increment
				Liquid hydrocarbon yield %	27.05	27.83	+0.77
Total liquid is received yield %	85.01	85.97	+0.96
				Total economic benefit (unit/ton raw material)	153.20	194.46	+41.26

From above test result, there is effect preferably in this system.

Embodiment 2: the online correlation integral optimal control of ketone benzene de-oiling dewaxing integrated unit

Ketone-benzol dewaxing device is the main device of refinery OIL IN LUBRICATING OIL PRODUCTION, and purpose is that the lubricating oil in the raw material is separated with paraffin, also is to belong to continuous flow procedure.The process chart of this process is as shown in figure 10:

Feedstock oil is divided into 7 the tunnel after entering system, every road is provided with 3 crystallizers, filtrate is at first added to dilute in advance to feedstock oil in each road, enter heat interchanger E101 afterwards, when feedstock oil enters E101, add fresh solvent and change filtrate after cold to carry out once (1) dilution, feedstock oil through once (1) dilution and through E101 change cold after, behind heat interchanger E101, add filtrate to carry out once (2) dilution, feedstock oil enters ammonia cooling crystallizer E102 afterwards, E103, add fresh solvent then to carry out secondary dilution, feedstock oil after the crystallization enters surge tank D101, the dewaxing filter, add cold wash solvent during this time, filtrate filtered enters solvent recovering system through filtrate tank D104.The wax that filters out enters wax flow container D105 and D106, through one section wax deoiling filter, during add one section cold wash, its filtrate enters filtrate tank D110.The wax that filters out of one section wax deoiling filter enters wax flow container D112, after the adding fresh solvent mixes with it, enters two sections wax deoiling filters and adds two sections cold wash.The filtrate of two sections wax deoiling filters enters filtrate tank D111, enters dewaxing filter wax flow container D105 then.The wax that two sections wax deoiling filters filter out mixes back formation with two sections diluting solvents that add in wax flow container D113 deoiled wax enters solvent recovering system.

The technical indicator that this system will optimize is the yield of wax of deoiling.And crucial performance variable (optimization variable) is the ratio of solvent that respectively flows each road, and is specific as follows:

The design optimization variable of optimal controller is 23, is respectively

1, one tunnel pre-dilution ratio

2, two tunnel pre-dilution ratios

3, three tunnel pre-dilution ratios

4, four tunnel pre-dilution ratios

5, five tunnel pre-dilution ratios

6, six tunnel pre-dilution ratios

7, seven tunnel pre-dilution ratios

8, one tunnel one rare (1) dilution ratio

9, two tunnel one rare (1) dilution ratio

10, three tunnel one rare (1) dilution ratio

11, four tunnel one rare (1) dilution ratio

12, five tunnel one rare (1) dilution ratio

13, six tunnel one rare (1) dilution ratio

14, seven tunnel one rare (1) dilution ratio

15, one tunnel one rare (2) dilution ratio

16, two tunnel one rare (2) dilution ratio

17, three tunnel one rare (2) dilution ratio

18, four tunnel one rare (2) dilution ratio

19, five tunnel one rare (2) dilution ratio

20, six tunnel one rare (2) dilution ratio

21, seven tunnel one rare (2) dilution ratio

22, secondary ratio

23, cold wash ratio

In this system, what the set-point control system was used is YOKOGAWA Centem CS system, and the optimal control computing machine is the active station of this system.The data window width of correlation integral optimal control is 13 hours.Gathered key operation condition and technical indicator (objective function: the data yield of the wax that deoils) every 60 seconds.Control calculating according to following steps:

1. name each key operation variable

u ₁=one tunnel pre-dilution ratio

u ₂=two tunnel pre-dilution ratios

u ₃=three tunnel pre-dilution ratios

u ₄=four tunnel pre-dilution ratios

u ₅=five tunnel pre-dilution ratios

u ₆=six tunnel pre-dilution ratios

u ₇=seven tunnel pre-dilution ratios

u ₈=one tunnel one rare (1) dilution ratio

u ₉=two tunnel one rare (1) dilution ratio

u ₁₀=three tunnel one rare (1) dilution ratio

u ₁₁=four tunnel one rare (1) dilution ratio

u ₁₂=five tunnel one rare (1) dilution ratio

u ₁₃=six tunnel one rare (1) dilution ratio

u ₁₄=seven tunnel one rare (1) dilution ratio

u ₁₅=one tunnel one rare (2) dilution ratio

u ₁₆=two tunnel one rare (2) dilution ratio

u ₁₇=three tunnel one rare (2) dilution ratio

u ₁₈=four tunnel one rare (2) dilution ratio

u ₁₉=five tunnel one rare (2) dilution ratio

u ₂₀=six tunnel one rare (2) dilution ratio

u ₂₁=seven tunnel one rare (2) dilution ratio

u ₂₂=secondary ratio

u ₂₃=cold wash ratio

The named target function:

J ₁The yield of=the wax that deoils

2. on active station, set up data acquisition system (DAS) with YOKOGAWA Centum CS Distributed Control System (DCS) and supporting data acquisition environment.And set up that to have the data window width be 13 hours real-time data base.This database was gathered the data of key operation condition and each technical indicator (each objective function) by YOKOGAWA Centem CS every 60 seconds.Data storage in the data window is whenever adopted sample one time among database, data window just moves forward a sampling time, that is to say, the oldest data are abandoned, and up-to-date data are added among the database.

3. after data sampling finishes, each operating conditions is carried out autocorrelation integral matrix k _UUCalculating: existing m=23 operating conditions

$k_{\mathrm{UU}}=\left\{\begin{array}{cccc}{k}_{11}& k_{12}& \·\·\·& k_{1m}\\ k_{21}& k_{22}& \·\·\·& k_{2m}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{m1}& k_{m2}& \·\·\·& k_{\mathrm{mm}}\end{array}\right\}$

In the formula: $k_{\mathrm{ij}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)u_j(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

I, j=1,2 ... m T=7200, M=14000 are integration constant, and this constant is to be determined by dewaxing process time constant roughly.

4. calculate the cross-correlation integral matrix K between each performance variable and the technical indicator _UJ:

K _UJ＝K _UJ1

In the formula:

$K_{\mathrm{UJ}1}=\left[\begin{array}{c}{k}_{u1J1}\\ k_{u2J1}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}1}\end{array}\right]$

$k_{\mathrm{uiJ}1}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)J_1(\mathrm{\λ}-\mathrm{\τ})\mathrm{d\λd\τ}$

(i＝1，2，…，m)

T, M are integration constant, T=7200, M=14000.

5. according to the autocorrelation integral matrix of above operating conditions and the cross-correlation integral vector between performance variable and each technical indicator, calculate operating conditions to the gradient vector between comprehensive technical indexes

Earlier obtain K by calculating following linear equation ^d:

K _UJ＝K _UUK ^d

And $K^d=K_{\mathrm{UJ}1}^d$

Integrated objective function J is to the gradient of operating conditions For

$K_{\mathrm{\σ}}^d={\mathrm{\σ}}_1{K}_{J1}^d$

In the formula:

σ ₁＝1

6. according to the gradient vector of resulting operating conditions to comprehensive technical indexes

Calculate the change direction of operating conditions, its principle is that then this operating conditions is at present at optimum state if the gradient of calculating is zero; If non-vanishing, carry out the operating conditions adjustment according to the size and the direction of gradient.For example: present operating conditions value is known, obtains adjusted operating conditions value according to the methods below, requires the technical indicator maximization:

In the formula:

$\left[\begin{array}{c}{u}_1\left(l\right)\\ u_2\left(l\right)\\ \·\\ \·\\ \·\\ u_m\left(l\right)\end{array}\right]$ Be original m operating conditions value, and $\left[\begin{array}{c}{u}_1(l+1)\\ u_2(l+1)\\ \·\\ \·\\ \·\\ u_m(l+1)\end{array}\right]$ Be adjusted new operating conditions.

$\left[\begin{array}{c}{k}_{1\mathrm{\σ}}^d\\ k_{2\mathrm{\σ}}^d\\ \·\\ \·\\ \·\\ k_{\mathrm{m\σ}}^d\end{array}\right]=K_{\mathrm{\σ}}^d$

α ₁, α ₂..., α _mBe m positive constant, the size of these constants is relevant with the speed of convergence of optimal control, need adjust at the scene.According to the adjustment result at scene, all value is 0.001.

7. got back to for the 3rd step.

We have obtained following table according to test:

Product yield before and after optimizing

	Pressed distillate (%)	Wax (%) deoils	Sweat oil (%)
				Before the optimization	50.13	42.64	7.67
After the optimization	49.57	43.85	6.9
				Increment	-0.56	+1.21	-0.77

From above result as can be seen, the technical indicator that optimize, the yield of the wax that deoils has increased by 1.21%, and certain optimization effect is arranged.

Claims (8)

1, a kind of real time operation optimizing method of multiple input and multiple output continuous producing process, it is characterized in that a plurality of key operation conditions in the production run as optimization variable, with the one or more technical indicators that are associated with the key operation condition is objective function, historical data according to production run key operation condition and technical indicator, online calculate between key operation condition and the technical indicator gradient vector in the current time, determine the adjustment direction of operating conditions then according to this gradient vector, when gradient vector be negative value or on the occasion of the time, all will adjust the key operation condition makes gradient vector change to zero direction, make technical indicator reach optimum, the calculating of this gradient is online ongoing, no matter whether technical indicator reaches optimum, in case find that gradient is non-vanishing, then adjusting the key operation condition makes gradient vector change to zero direction, to realize the tracking of optimum point, the method that online calculates between key operation condition and the technical indicator in the gradient vector of current time is to utilize the correlation integral technology that the historical data of production run key operation condition and technical indicator is carried out integral operation, and described correlation integral technology comprises:

The structure objective function (t), this objective function should be online computable or online measurable, so $\tilde{J}\left(t\right)=f(\tilde{u}\left(t\right),\tilde{p}\left(t\right),t)$ , wherein,

(t) be m dimension average optimized controllable variable,

(t) for disturbing, f is a unknown mappings;

According to formula

$k_{\mathrm{uu}}=\left[\begin{array}{cccc}{k}_{u1u1}& k_{u1u2}& \·\·\·& k_{u1\mathrm{um}}\\ k_{u2u1}& k_{u2u2}& \·\·\·& k_{u2\mathrm{um}}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{\mathrm{umu}1}& k_{\mathrm{umu}2}& \·\·\·& k_{\mathrm{umum}}\end{array}\right]$

$k_{\mathrm{uiuj}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})u_j\left(t\right)\mathrm{dtd\τ}$ Real-time measurement values u with optimization variable _i(t) (i=1,2 ..., m) the autocorrelation integral matrix k of calculation optimization variable _UuT wherein, M is the integration constant greater than 0;

According to formula

$k_{\mathrm{uJ}}=\left[\begin{array}{c}{k}_{u1J}\\ k_{u2J}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}}\end{array}\right]$

$k_{\mathrm{uiJ}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i(t-\mathrm{\τ})J\left(t\right)\mathrm{dtd\τ},(i=\mathrm{1,2},\·\·\·,m)$ With real-time measurement values J (t) the calculation optimization variable of objective function and the cross-correlation integral vector of target variable; T wherein, M is the integration constant greater than 0;

According to formula $k_{\mathrm{uJ}}=k_{\mathrm{uu}}\frac{\mathrm{dEf}}{\mathrm{dE}\left\{\tilde{u}\left(t\right)\right\}}+\mathrm{\ϵ}\left(t\right)$ Obtain the gradient of objective function

According to formula $u_s(l+1)=u_s\left(l\right)+\mathrm{\α}\frac{\mathrm{dEf}}{\mathrm{dE}\left\{\tilde{u}\left(l\right)\right\}}$ The new settings value us (l+1) of calculation optimization variable; In the formula, α is a constant.

2, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 1 is characterized in that may further comprise the steps:

(1), determines that a plurality of technical indicators that will optimize are objective function J according to the needs of optimised process ₁, J ₂..., J _n, these objective functions must onlinely can calculate or measure, and construct a comprehensive objective function J=σ ₁J ₁+ σ ₂J ₂+ ... ,+σ _nJ _n, σ here ₁, σ ₂..., σ _nFor the weighted number of each objective function, according to technological requirement, value is between 0 to 1, and weighted value is time dependent here,

Definite key operation condition u that will optimize ₁, u ₂..., u _mAs a plurality of variablees that will optimize;

The key operation condition that will optimize is carried out conventional definite value control, and its setting value adopts the correlation integral technology to calculate by the optimal control computing machine, carries out the adjustment of a setting value every cycle regular hour;

(2) set up real-time data acquisition system, gather the data of key operation condition and technical indicator, to obtain the historical data of production run key operation condition and technical indicator with data window;

(3) after data sampling finishes, each key operation condition is carried out autocorrelation integral matrix k _UUCalculating: be provided with m operating conditions, then

$k_{\mathrm{UU}}=\left\{\begin{array}{cccc}{k}_{11}& k_{12}& \·\·\·& k_{1m}\\ k_{21}& k_{22}& \·\·\·& k_{2m}\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ \·& \·& \·& \·\\ k_{m1}& k_{m2}& \·\·\·& k_{\mathrm{mm}}\end{array}\right\}$

In the formula: $k_{\mathrm{ij}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)u_j(\mathrm{\λ}-\mathrm{\τ})\mathrm{dtd\τ}$

I, j=1,2 ... m T, M are integration constant,

The information that in the autocorrelation integral matrix, has comprised correlativity between each operating conditions,

(4) calculate cross-correlation integral matrix K between each performance variable and the technical indicator _UJ: be provided with n objective function, then

$K_{\mathrm{UJ}}=\left[\begin{array}{c}{K}_{\mathrm{UJ}1}\\ K_{\mathrm{UJ}2}\\ \·\\ \·\\ \·\\ K_{\mathrm{UJn}}\end{array}\right]$

In the formula:

$K_{\mathrm{UJ}1}=\left[\begin{array}{c}{k}_{u1J1}\\ k_{u2J1}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJ}1}\end{array}\right]......K_{\mathrm{UJn}}=\left[\begin{array}{c}{k}_{u1\mathrm{Jn}}\\ k_{u2\mathrm{Jn}}\\ \·\\ \·\\ \·\\ k_{\mathrm{umJn}}\end{array}\right]$

$k_{\mathrm{UiJs}}={\∫}_{-M}^M\frac{1}{2T}{\∫}_{-T}^T{u}_i\left(\mathrm{\λ}\right)J_s(\mathrm{\λ}-\mathrm{\τ})\mathrm{dtd\τ}$

i＝1，2，…，m；s＝1，2，…，n

T, M are integration constant;

(5) according to the autocorrelation integral matrix of above operating conditions and the cross-correlation integral vector between performance variable and each technical indicator, calculate operating conditions to the gradient vector between comprehensive technical indexes

Earlier obtain K by calculating following linear equation ^d:

K _UJ＝K _UUK ^d

And

$K^d=\left[\begin{array}{c}{K}_{J1}^d\\ K_{J2}^d\\ \·\\ \·\\ \·\\ K_{\mathrm{Jn}}^d\end{array}\right]$

Integrated objective function J is to the gradient of operating conditions

For

$K_{\mathrm{\σ}}^d={\mathrm{\σ}}_1{K}_{J1}^d+{\mathrm{\σ}}_2{K}_{J2}^d,\·\·\·,{\mathrm{\σ}}_n{K}_{\mathrm{Jn}}^d;$

(6) according to the gradient vector of resulting operating conditions to comprehensive technical indexes

Calculate the change direction of operating conditions, its principle is that then this operating conditions is at present at optimum state if the gradient of calculating is zero; If non-vanishing, carry out the operating conditions adjustment according to the size and the direction of gradient;

(7) adjust finish after, at regular intervals, carry out data sampling once more, return step 3,

Step 3 to 6 process is online ongoing, makes each operating point finally reach optimum point.

3, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 2, it is characterized in that in the step (1) the key operation condition that will optimize being carried out conventional definite value control is to be that DCS or conventional instrument carry out conventional definite value control to the operating conditions that will optimize by the distributing system computing machine earlier, and its setting value adopts the correlation integral technology to calculate by the optimal control computing machine, carries out the adjustment of a setting value every cycle regular hour.

4, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 2, it is characterized in that in the step (2), be according to concrete process time characteristic, foundation has the real-time data acquisition system of data window, the method of gathering key operation condition and technology index data is to set up the real-time data acquisition data system with certain data window width, this system is that DCS constitutes by Distributed Control System (DCS), to obtain the historical data of production run key operation condition and technical indicator, it is the data of each objective function with each technical indicator that the key operation condition is gathered every certain sampling interval by this system, data storage in the data window is among database, whenever adopt sample one time, data window just moves forward a sampling time, the oldest data are abandoned, and up-to-date data are added among the database.

5, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 4 is characterized in that in the step (2), the time width of the data window width of foundation greater than optimization variable to more than 3 times of objective function settling time.

6, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 4, it is characterized in that in the step (2) that real-time data acquisition data system collection key operation condition and each technical indicator are that the data of each objective function are to gather once second every 30--90 according to the speed of process.

7, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 2 is characterized in that in the step (6), and the gradient of calculating is zero, and then this operating conditions is at present at optimum state; The gradient of calculating is non-vanishing, carries out the operating conditions adjustment according to the size and the direction of gradient, and be known in the operating conditions value, obtain adjusted operating conditions value according to the methods below, require the comprehensive technical indexes maximization:

In the formula:

$\left[\begin{array}{c}{u}_1\left(l\right)\\ u_2\left(l\right)\\ \·\\ \·\\ \·\\ u_m\left(l\right)\end{array}\right]$ Be original m operating conditions value, and $\left[\begin{array}{c}{u}_1(l+1)\\ u_2(l+1)\\ \·\\ \·\\ \·\\ u_m(l+1)\end{array}\right]$ Be adjusted new operating conditions. $\left[\begin{array}{c}{k}_{1\mathrm{\σ}}^d\\ k_{2\mathrm{\σ}}^d\\ \·\\ \·\\ \·\\ k_{\mathrm{m\σ}}^d\end{array}\right]=K_{\mathrm{\σ}}^d$

If ask technical indicator maximal value, then α ₁, α ₂..., α _mBe m positive constant.

8, the real time operation optimizing method of multiple input and multiple output continuous producing process according to claim 5 is characterized in that in the step (6), to i operating conditions u _i, adjust step-length at every turn and be Get suitable α _iValue, if promptly ask the maximal value of technical indicator, get α _iFor on the occasion of, otherwise get negative value, to adjust the size of step-length.

Images