Disclosure of Invention
The invention aims to provide a multi-variable strip shape control method of a cold rolling mill based on a constraint optimization algorithm, which has the advantages of high calculation efficiency, saving storage space of a controller, capability of parallel calculation, high control precision and strong system operation stability.
In order to realize the purpose, the following technical scheme is adopted: the control method comprises the steps of formulating an optimization model for solving the optimal regulating quantity of each plate shape regulating mechanism through mathematical modeling, and designing a multivariable optimization algorithm with global convergence around the optimization model; the multivariable optimization algorithm converts a multivariable plate-shaped optimization problem with constraint into a series of univariate optimization problems which can be searched along the coordinate direction; a step length acceleration method is designed, and through an iterative process, the target function can be quickly reduced, so that the optimal adjustment quantity of each plate shape adjusting mechanism is obtained;
the method comprises the following specific steps:
step 1, designing a multivariable optimization model for solving the optimal regulating quantity of a plate shape regulating mechanism for bending a working roll, bending a middle roll, inclining a roll and transversely moving the middle roll; adopting the weighted square sum of the residual plate shape deviation as an objective function of an optimization model; the mechanical design limit of each plate shape adjusting mechanism is used as a constraint condition;
step 2, converting a multivariable optimization model for solving the optimal adjustment quantity of the plate-shaped adjusting mechanism into a standard form of an optimization problem;
step 3, designing an initial search point for solving a standard form multivariable optimization problem;
step 4, designing a search mechanism along each coordinate axis direction;
step 5, searching along each coordinate axis direction, and calculating a candidate value of an optimal point;
step 6, designing a search termination criterion;
and 7, establishing a multi-variable strip shape control model of the cold rolling mill based on a constraint optimization algorithm.
Further, in step 1, the mechanical design limit of each plate shape adjusting mechanism is used as a constraint condition, and the optimization model is as follows:
wherein J (Δ u) is an objective function of the optimization model; delta u is the adjustment vector of each plate shape adjusting mechanism to be solved, and delta u belongs to R n (ii) a m is the number of strip shape measuring sections in the width direction of the strip; n is the number of the plate-shaped adjusting mechanisms; i and j are respectively the serial number of the measuring section and the serial number of the adjusting mechanism; g i Weighting factors for the plate shape deviation of the ith measurement section; Δ y i Is the plate shape deviation of the ith measurement section; Δ u j Is the adjustment quantity of the jth plate-shaped adjusting mechanism to be solved; eff ij The shape regulation efficiency coefficient of the jth plate shape regulating mechanism to the ith measuring section; BL j And BU j Respectively designing a lower limit and an upper limit of the mechanical design of the jth plate-shaped adjusting mechanism; u. of j Is the set value of the current control period of the jth plate shape adjusting mechanism.
Further, in step 2, the multivariate optimization model after being converted into the standard form is:
wherein f (x) is an objective function of the optimization model; x is the optimal regulating quantity vector of each plate-shaped regulating mechanism to be solved; Ω is an allowable set of constraints; l and u are respectively the lower limit and the upper limit of the optimal adjustment quantity of each plate shape adjusting mechanism.
Further, the specific method of step 3 is as follows: selecting an initial point x 0 E.g. omega, initial step length t 0 E (0, infinity), the initial search direction is any coordinate axis direction e i And i ∈ {1,..., n }; let the loop count variable initially be 1, i.e.: k =1,k ∈ { 1., K }, K = (1., ∞), K is a threshold for the total number of cycles, and three search termination thresholds ∈ are set 1 ,ε 2 And ε 3 ∈(0,1)。
Further, the specific method of step 4 is as follows: first along the coordinate axis e 1 Maximum step size t of direction search max And x is k +t max ·e 1 E is omega; from x k-1 At a step size of t = min { t } 0 ,t max Along a coordinate axis e 1 Starts to search in the positive direction to obtain a new point x after the first search k (11) =x k-1 +t·e 1 (ii) a If there is t>, 0 and f (x) k (11) )<f(x k-1 ) Let t = min { t } 0 /δ,t max Accelerating the search in the direction until the constraint boundary point or the objective function value is found to be decreased; if t = t max Or f (x) k (11) )>f(x k-1 ) Returning to the previous searching point and taking the point as an end point along the searching direction; if along the coordinate axis e 1 If the forward search of (2) does not decrease the objective function value, the coordinate axis e is set 1 The negative direction of (2) as the search direction; in this case the maximum search step in that direction needs to be recalculated, i.e.: t is t max And x k -t max ·e 1 E Ω, the search step then becomes t = min { t } 0 ,t max }; according to the formula e 1 The same search mode is searched in the positive direction to obtain the search end point x in the direction k (1) (ii) a If along the coordinate axis e 1 If the value of the objective function is not reduced by the positive and negative direction searches, the initial search point x is obtained k-1 Will be the end point x searched along the coordinate axis k (1) (ii) a Similarly, the search for other coordinate axis directions is performed according to this mechanism.
Further, the specific method of step 5 is as follows: first along the coordinate axis e 2 Maximum step size t of direction search max And x is k +t max ·e 2 E is omega; from x k (1) At a step size of t = min { t } 0 ,t max Along a coordinate axis e 2 Starts to search in the positive or negative direction to obtain the coordinate axis e 2 End point x of k (1) (ii) a By parity of reasoning, the third coordinate axis direction e is completed 3 To the nth coordinate axis direction e n The candidate optimal point x after the k searches are completed can be obtained k (n) (ii) a At this time, let x k =x k (n) (ii) a If not | | x k -x k-1 ||<ε 1 And if the iteration number K does not exceed the iteration number threshold K, making K = K +1, and according to the mechanism designed in the step four, selecting the coordinate direction e 1 Starting to the nth coordinate axis direction e n Restarting a new round of search; if the iteration number K is equal to the set iteration number threshold value K, x k Will be the optimal solution x to the optimization problem in step 2 * 。
Further, the specific method of step 6 is as follows: judging the termination of the searching process, wherein the judgment criterion comprises the checking of the searching step length and the variation of the objective function value; if the search step t 0 Is already small enough, i.e. satisfies t 0 <ε 2 (ii) a Or the variation of the objective function value is small enough, namely, | | f (x) is satisfied k )-f(x k-1 )||<ε 3 Then x at this time k Will be the optimal solution x to the optimization problem in step 2 * (ii) a Otherwise, shrink the search step length, let t 0 =θ·t 0 Then according to the searching mechanism designed in step 4, the coordinate direction e is used 1 Starting in the direction of coordinate axes e n A new round of search is restarted.
Further, the specific method in step 7 is as follows: establishing a multivariable strip shape closed-loop control model according to the constrained multivariable global optimization method established in the steps 3-6; and calculating the optimal regulating quantity of each plate shape regulating mechanism in each control period on line by a formulated global optimization method to be used as an input value of each plate shape regulating mechanism controller.
Compared with the prior art, the invention has the following advantages:
1. the global optimization method adopting a non-derivation mode can avoid solving complex objective function gradient information, does not need to construct a complex model based on the gradient information, and is beneficial to actual engineering development.
2. The method converts the multivariable strip shape optimization model into a series of univariate optimization problems, has the characteristics of high calculation efficiency, saving of storage space of a controller and capability of parallel calculation, and is very suitable for development of a cold-rolled strip shape control system with a plurality of strip shape adjusting mechanisms.
3. The control method has the advantages of high calculation precision, good universality, high plate shape control precision, strong system operation stability and the like after the detection of field practical application.
example 1: this example 1 discloses the process of multivariable strip shape optimization algorithm in the final frame plate shape closed-loop control system of a 1450mm five-stand cold continuous rolling mill. The plate shape adjusting mechanism comprises a roll inclination, a working roll positive/negative bending roll, a middle roll positive bending roll and a middle roll transverse moving, and main control parameters and rolling parameters are shown in a table 1.
TABLE 1 Main parameters of the Rolling Mill
The control method comprises the steps of formulating an optimization model for solving the optimal regulating quantity of each plate shape regulating mechanism through mathematical modeling, and designing a multivariable optimization algorithm with global convergence around the optimization model; the multivariable optimization algorithm converts the multivariable plate shape optimization problem with constraints into a series of univariate optimization problems which can be searched along the coordinate direction; a step length acceleration method is designed, and the target function can be quickly lowered through an iterative process, so that the optimal adjustment quantity of each plate shape adjusting mechanism is obtained;
as shown in fig. 4, the specific steps are as follows:
step 1, designing a multivariable optimization model for solving the optimal regulating quantity of the plate shape regulating mechanism of the working roll bending roll, the middle roll bending roll, the roll inclination and the middle roll transverse moving. The establishment of the optimization model comprises three basic elements which are respectively used for determining a complete decision variable, a clear objective function expression and a constraint condition. For multivariable strip shape control, the adjustment quantity of the strip shape adjusting mechanism to be solved and the corresponding value interval thereof can be used as decision variables and constraint conditions of an optimization model. And the target of the strip shape control is to eliminate the residual strip shape deviation to the maximum extent, so that the strip shape deviation remained after the action of each strip shape adjusting mechanism is taken as the target function of the optimization model. The smaller the residual plate shape deviation is, the better the adjustment quantity of the plate shape adjusting mechanism is. It is considered that whether the objective function is a univariate function or a multivariate function, it must be a convex function or only have a peak to guarantee a globally optimal solution. The weighted sum of squares of the residual plate shape deviation is therefore used as the objective function of the optimization model. The optimization model for calculating the optimum adjustment for each strip shape adjustment mechanism can be expressed as:
wherein J (Δ u) is an objective function of the optimization model; delta u is the adjustment vector of each plate shape adjusting mechanism to be solved, and delta u belongs to R n (ii) a m is the number of strip shape measuring sections in the width direction of the strip; n is the number of the plate-shaped adjusting mechanisms; i and j are respectively a measuring section serial number and an adjusting mechanism serial number; g i Weighting factors for the plate shape deviation of the ith measurement section; Δ y i Is the plate shape deviation of the ith measurement section; Δ u j Is the adjustment quantity of the jth plate-shaped adjusting mechanism to be solved; eff ij The adjustment efficiency coefficient of the jth plate shape adjusting mechanism to the plate shape of the ith measuring section is obtained; BL j And BU j Respectively designing a lower limit and an upper limit of the mechanical design of the jth plate-shaped adjusting mechanism; u. of j Is the set value of the current control period of the jth plate shape adjusting mechanism. The shape multivariable optimization model is established based on the shape regulation efficacy, so that the shape regulation efficacy coefficient of each shape regulation mechanism needs to be calculated and determined. The strip shape regulation efficiency coefficient represents the variation of a strip shape value caused by a strip shape regulating mechanism under the condition of generating unit regulating quantity, and the calculation method is as follows:
wherein Eff is a matrix of the plate shape regulation efficacy coefficient, and Eff is in the same size as R m×n (ii) a U is the adjustment vector of each plate shape adjusting mechanism, and delta U belongs to R n (ii) a Δ Y is the amount of change in shape of the induced plate, and Δ Y ∈ R m 。
FIG. 1 shows the calculated profile shape control efficiency coefficient curves of the profile shape adjusting mechanisms such as the work roll bending, the middle roll transverse moving and the roll inclination.
Since the final stand of the 1450mm five stand cold continuous mill was the UCM mill, there were four plate shape adjustment mechanisms in total, so the number of adjustment mechanisms in this example model was 4, i.e., n =4. In order to simplify the data processing process of the control system, the actual number of measurement sections and the target plate shape and the plate shape deviation at each measurement section are equivalent to the plate shape target value and the plate shape deviation value at 20 measurement sections, that is, m in the model is 20. The remaining parameters of the model in this example are shown in table 2.
TABLE 2 principal parameters of the multivariate strip shape optimization model
Mechanical design positive limit BU of the individual plate shape adjusting mechanisms in this example j Negative limit BL j Set to 100% and-100%, respectively. 100% represents that the actuator has reached a positive adjustment limit, and 100% represents that a negative adjustment limit has been reached. The adjustment, the set value and the actual value of each profile adjustment means are defined as percentages of the mechanical limit. The strip shape value and the strip shape deviation value represent 10 of the elongation 5 The power factor is dimensionless like the power factor of the shape control.
Step 2, converting a multivariable optimization model for solving the optimal adjustment quantity of the plate-shaped adjusting mechanism into a standard form of an optimization problem;
to introduce the multivariate optimization algorithm more clearly, the multivariate strip optimization model is transformed into a standard form of the multivariate optimization problem. The multivariate optimization model after conversion to the standard form is:
wherein f (x) is an objective function of the optimization model; x is the optimal regulating quantity vector of each plate-shaped regulating mechanism to be solved; Ω is an allowable set of constraints; l and u are respectively the lower limit and the upper limit of the optimal adjustment quantity of each plate shape adjusting mechanism. These parameters can all be derived from the data given in step 1. For this optimization problem, if the allowed point x is not a stagnation point of the function f (x), then there must be one allowed y to satisfyk belongs to {1,. Eta., n }. If t = (y-x) k >, 0, and considering Ω is the box constraint, then there are:
in the formula:is the gradient of the objective function f (x) at coordinate x; e.g. of the type k is an orthogonal set of coordinate directions, and k ∈ {1,..., n }.
Due to the fact thatWith continuity, and the allowed set Ω is a convex set, there must be a positive value λ satisfying:
f(x+t·e k )<f(x),x+t·e k ∈Ω,t∈(0,λ)
when t = (y-x) k &When lt is 0, it is represented by-e k Instead of e k . Thus, for any allowed point x, even if it is notIs the function stagnation point, but a coordinate direction exists necessarily to enable the objective function to have strict decline. This direction may be a positive direction of the coordinate axis or a negative direction of the coordinate axis.
Step 3, designing and solving an initial search point of a standard form multivariable optimization problem; selecting an initial point x 0 E.g. omega, initial step length t 0 E (0, infinity), the initial search direction being any coordinate axis direction e i And i ∈ {1,..., n }; let the loop count variable initially be 1, i.e.: k =1,k ∈ { 1., K }, K = (1., ∞), K is a threshold for the total number of cycles, and three search termination thresholds ∈ are set 1 ,ε 2 And ε 3 ∈(0,1)。
In this embodiment 1, the initial point is selected as x 0 = (0.5,0.5,0.5,0.5,0.5) and f (x) 0 ) =57.32483, initial step t 0 =1, initial search direction e 1 =1,0,0,0, the step-size reduction factor and the step-size acceleration factor are δ = θ =0.5, respectively, the threshold K =4000 for the total number of cycles, and three search termination thresholds are set to be ∈ respectively 1 =10 -6 ,ε 2 =10 -6 And ε 3 =10 -7 。
Step 4, designing a search mechanism along each coordinate axis direction; first along the coordinate axis e 1 Maximum step size t of direction search max =0.5, then from x 0 = (0.5,0.5,0.5,0.5,0.5) at step t = min {1,0.5} along coordinate axis e 1 Starts to search in the positive direction, and obtains a new point x after the first search 1 (11) =x 0 +t·e 1 . If there is t>, 0 and f (x) 1 (11) )<f(x 0 ) Let t = min { t } 0 /0.5,t max I.e. speeding up the search in that direction until the constraint boundary point is searched for or the value of the objective function falls again. If t = t max Or f (x) 1 (11) )>f(x 0 ) Then return to the previous search point and take that point as the end point along the search direction. If along the coordinate axis e 1 If the positive direction search can not make the objective function value decrease, the coordinate axis e is set 1 In the negative direction ofFor the search direction, and recalculate the search step size according to 1 The same search mode is searched in the positive direction of the direction to obtain the search end point x of the direction k (1) . Similarly, the search for other coordinate axis directions is performed according to this mechanism.
Step 5, searching along each coordinate axis direction according to the searching mechanism designed in the step 4, calculating a candidate value of an optimal point, and determining the optimal point of the objective function according to a search termination criterion;
first along the coordinate axis e 2 Maximum step size t of direction search max And x is k +t max ·e 2 E is omega; from x k (1) At step size t = min { t } 0 ,t max Along a coordinate axis e 2 Starts to search in the positive or negative direction to obtain the coordinate axis e 2 End point x of k (1) (ii) a By parity of reasoning, the third coordinate axis direction e is completed 3 To the nth coordinate axis direction e n The candidate optimal point x after the k searches are completed can be obtained k (n) (ii) a At this time, let x k =x k (n) (ii) a If not | | x k -x k-1 ||<ε 1 And if the iteration number K does not exceed the iteration number threshold K, making K = K +1, and according to the mechanism designed in the step four, selecting the coordinate direction e 1 Starting to the nth coordinate axis direction e n Restarting a new round of search; if the iteration number K is equal to the set iteration number threshold value K, x k Will be the optimal solution x to the optimization problem in step 2 * . The trajectory of the drop of the objective function value during the optimization is shown in fig. 5, and the number of iterations and the main results are shown in table 3.
TABLE 3 model optimization results
After the steps 4 and 5 are completed, search iteration in each coordinate axis direction can be completed, and the specific algorithm structures of the steps 4 and 5 are shown in fig. 2.
Step 6, designing a search termination criterion; judging the termination of the searching process, wherein the judgment criterion comprises the check of the searching step length and the variable quantity of the objective function value; if the search step t 0 Is already small enough, i.e. satisfies t 0 <ε 2 (ii) a Or the variation of the objective function value is small enough, namely, | | f (x) is satisfied k )-f(x k-1 )||<ε 3 Then x at this time k Will be the optimal solution x of the optimization problem in step 2 * (ii) a Otherwise, shrink the search step length, let t 0 =θ·t 0 Then according to the searching mechanism designed in step 4, the coordinate direction e is used 1 Starting in the direction of coordinate axes e n A new round of search is restarted. If the algorithm generates a sequence of iterations { x } k Converge to a limit point, each limit point will be the stagnation point of the multivariate optimization problem, or an approximation of the stagnation point.
And 7, establishing a multi-variable strip shape control model of the cold rolling mill based on a constraint optimization algorithm. Establishing a multivariable strip shape closed-loop control model according to the constrained multivariable global optimization method established in the steps 3-6; the optimal adjustment quantity of each strip shape adjusting mechanism in each control period is calculated on line through a formulated global optimization method and is used as an input value of each strip shape adjusting mechanism controller, as shown in fig. 3. In fig. 3, Δ Y represents a deviation vector between the target strip shape and the measured strip shape; eff represents a plate shape regulation efficiency coefficient matrix of the plate shape regulating mechanism; Δ U = { Δ U tr ,Δu wrb ,Δu irb ,Δu irs Expressing the optimal adjustment amount of each strip shape adjustment mechanism calculated by a multivariate optimization model; Δ u tr ,Δu wrb ,Δu irb ,Δu irs Respectively representing the optimal adjustment amounts of roll inclination, work roll bending, intermediate roll bending and intermediate roll transverse movement; u = { U = tr ,u wrb ,u irb ,Δu irs Representing the actual value vector of each plate shape adjusting mechanism in the current control period; p represents the vector of the set value of each strip shape adjusting mechanism in the next control cycle.
According to the strip shape deviation and the determined strip shape regulation and control efficiency coefficient, the optimal regulating quantity of each strip shape regulating mechanism in each control period is calculated on line through a formulated global optimization method and is used as an input value of each strip shape regulating mechanism controller, and therefore the strip shape of the rolling mill is continuously, dynamically and in real time controlled, and a multivariable strip shape closed-loop control system of a 1450mm five-stand cold continuous rolling mill formulated in the embodiment is shown in fig. 6.
FIG. 7 is a set of profile target values and profile measurement values of the strip along the width direction of the strip after the optimization model is put into operation. As can be seen from the data analysis, the measured value of the strip shape is basically consistent with the target value, the deviation of the strip shape at each measuring section is very small, and the deviation is basically within +/-2I-Unit, so that the strip shape control precision is higher.
FIG. 8 shows the CPU load profile during rolling (model SIMATIC TDC CPU551, available from Siemens). As can be seen from the figure, the designed multivariable plate shape optimization algorithm does not occupy excessive system resources, and the CPU is in a lower stable operation interval.
The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements of the technical solution of the present invention by those skilled in the art should fall within the protection scope defined by the claims of the present invention without departing from the spirit of the present invention.