CN1110729C - Golden-section intelligent control method based on description of object characteristic-model - Google Patents

Abstract

The present invention relates to a golden section intelligent control method based on the description of an object characteristic model, which aims to solve two key problems in the design of an intelligent controller. The present invention provides a reasonable effective and simple model for the design of the intelligent controller, and simultaneously provides a design method for the intelligent controller which has the advantages of high performance, strong robustness, simplicity and practicality.

Description

Golden-section intelligent control method based on the characteristics of objects model description

The invention belongs to field of intelligent control technology, a kind of golden-section intelligent control method of saying so exactly based on the characteristics of objects model description.

Traditional control method, comprise the classical way described based on transport function and based on the modernism of state space description, be too dependent on accurate modeling, therefore often has significant limitation in actual applications, and the structure of spacecrafts such as following applied satellite, space station becomes increasingly complex (lightweight, flexible, many bodies, liquid sloshing and mass change etc.), add unknown or uncertain variation of structure, shape and external environment, make accurate modeling very difficult.Meanwhile, the requirement of control task but more and more higher (high precision, high stable, highly reliable etc.), this only depends on traditional control method to run into difficulty with regard to making.

Be difficult to the severe challenge that requires in the face of the controlled device that becomes increasingly complex and more and more higher control just because of traditional control method, be subjected to increasing people's attention before and after the seventies by the Based Intelligent Control of some scholars' propositions of the U.S..But some intelligence control methods that occurred at present (passing rank Based Intelligent Control, expert's control etc. as fuzzy control, ANN (Artificial Neural Network) Control, layering), also lack the complete thought of carrying out the Based Intelligent Control design in advance, great majority all do not utilize Properties of Objects well, the foundation that lacks CONTROLLER DESIGN, therefore main structure and the parameter of determining controller by on-the-spot " examination is gathered ".And, can not satisfy required control performance index request because " examination is gathered " often causes the transient performance the unloading phase of control bad, go thereby be difficult to promote the use of in the working control engineering.For example to the control of spacecraft model, it is perfectly safe to require to accomplish, once success can not be leaned on " examination is gathered ".

At US Government Research Report B88-150610 " adaptive control algorithm of simplification) " (Oxford Univ. (England) .Dept.of Engineering Science.Simplified Self-tuning ControlAlgorithms, OUEL-1657/86,1986,65p, warwick, k.) in, in order to give simple adaptive controller of high-order object designs, at first estimate the kinetic model parameter of object, the application model reducing technique obtains a simple model then, again according to this simple modelling PID controller.This controller design method of being proposed of report has utilized typical control device design philosophy, and its shortcoming is: (1) is uncertain of or pumping signal inadequately fully the time when the exponent number of object model, and parameter estimation might not converge to true value; (2) model reduction can cause that observation overflows or control overflow problem; (3) utilize simple PID controller, system performance and robustness thereof that whole closed-loop control system can reach are limited.

The object of the present invention is to provide a kind of golden-section intelligent control method based on the characteristics of objects model description, to solve two key issues in the Design of intelligent controller, be design of Intelligence Controller a rational and effective naive model is provided, the method for designing of high-performance, strong robustness, simple, Practical Intelligent controller is provided simultaneously.

To achieve these goals, the present invention has adopted following steps:

1. the feature of controlled device is at first analyzed in modeling, promptly whether belong to linear stational system and can be linearizing the nonlinear constant system; Analyze the control performance requirement that above-mentioned object satisfies position maintenance and Position Tracking Control then, the difference equation of deformation type and four characteristic models that characteristic parameter is formed when requiring to set up a second order with control according to above-mentioned characteristics of objects;

2. characteristic model is carried out parameter estimation, (1) determines parameter estimation initial value and parameter estimation scope according to overall coefficient self-adaptation control method principle; (2) in the transient process stage, utilize the output full dose to set up full dose parameter estimation formula, the parameter of characteristic model is estimated, and estimated result rectangular projection is arrived in the parameter estimation scope of determining the front; (3) in the control of position constant, as enter steady-state zone and output error when bigger than setting value, utilize output error to set up margin of error parameter estimation formula, the parameter of characteristic model is estimated, and estimated result rectangular projection is arrived in the parameter estimation scope of determining the front;

3. carry out Design of intelligent controller, (1) design is kept and tracking control unit; (2) design golden section feedback controller; (3) design logic integration and logic derivative controller;

4. set up three class knowledge bases (dictionary, Runtime Library and buffer pool).Carry out the The whole control system software design.

The golden section controller that the present invention proposes can solve the start-up period of the bad difficulty of transient process quality general classic method is controlled at to(for) the unknown parameters object; The logic integral controller and the logic derivative controller that propose have intelligent characteristics, aspect the adverse effect of resisting uncertain process interference and measurement noise, are far superior to general traditional integration, derivative compensation device.

The application of the present invention in Spacecraft Control shortened the lead time for the precision, reliability and the antijamming capability that improve attitude of satellite control, saves research fund, and the development that promotes Theory of Automatic Control, method and technology is extremely important.

Except that space industry, the present invention has great application value in Industry Control, all can use as fields such as oil, chemical industry, metallurgy, electric power.

Fig. 1 is an intelligence control system structured flowchart of the present invention;

Fig. 2 is total flow chart of the present invention.

Below in conjunction with accompanying drawing the present invention is further specified:

As shown in Figure 1, mainly be made up of six parts based on intelligence control system of the present invention: 1. characteristic model is described and the characteristic parameter estimation; 2. Design of intelligent controller; 3. knowledge base and knowledge are revised; 4. intelligent management; 5. reference model; 6. controlled device and measurement component and execution unit accordingly.

The present invention mainly is 1., 2., 3. aspect three, comprise 1. that wherein characteristic model is described and two parts of characteristic parameter estimation.Below just revise four parts and specifically be described below by characteristic model description, characteristic parameter estimation, Design of intelligent controller, knowledge base and knowledge: 1. the characteristic model of controlled device is described

Variation is divided equation when (1) setting up a second order

High-order object for the permanent unknown parameters of linearity (known also can) is found the solution linearizing object G (s), under constant, Position Tracking or second order tracking Control situation, sets up the characteristic model equation of controlled device, the second order difference equation that becomes when promptly setting up:

y(k)＝f ₁(k)y(k-1)+f ₂(k)y(k-2)+g ₀(k)u(k-k ₀-1)+g ₁(k)u(k-k ₀-2)

=φ ^T(k) θ (k) (1) φ (k)=[y (k-1) y (k-2) u (k-k wherein ₀-1) u (k-k ₀-2) ^T,

θ(k)＝[f ₁(k)?f ₂(k)?g ₀(k)?g ₁(k)] ^T，

Y (k) is system's output quantity, and u (k) is input controlled quentity controlled variable, k ₀Be the hysteresis step number.For minimum phase system, can adopt simpler mode modeling, even

Y (k)=φ ^T(k) θ (k) (2) φ (k)=[y (k-1) y (k-2) u (k-k wherein ₀-1)] ^T

θ (k)=[f ₁(k) f ₂(k) g ₀(k)] ^TEngineering with generally all can be by formula (2) modeling.(2) determine four characteristic parameters

1. determine or estimate the minimum equivalent time constant T of controlled device _Min

2. determine or estimate the scope of the static gain D or the estimation static gain D of controlled device

3. select rational time in sampling period Δ t, $\frac{\mathrm{\Δt}}{T_{\min }}=\frac{1}{5}~\frac{1}{15}$

4. measure pure retardation time of τ, determine or estimate the pure hysteresis step number k of controlled device ₀, general calculating $k_0=\frac{\mathrm{\τ}}{\mathrm{\Δt}}$ Round numbers, some explanation is arranged when engineering is used: to controlled device output y (k) and controlled quentity controlled variable u (k) as the PID control instrument to carry out the nominal processing like that, no matter output y (k) and input controlled quentity controlled variable u (k) are any physical units, y when for example controlling temperature (k) is degree, and the control attitude angle is an angle position etc., and all turning to unified nominal value is that full scale is 100%.2. the parameter estimation of characteristic model

(1) determines the parameter estimation scope

1.4331≤f ₁(k)≤2，

-1≤f ₂(k)≤-0.5134，

0.9196≤f ₁(k)+f ₂(k)≤1，

g ₀(k) ∈ [0.003 0.3] minimum phase system,

g ₀(k) ∈ [0.001 0.01] non-minimum phase system,

g ₁(k) ∈ [0.003 0.03] non-minimum phase system.

(2) determine initial parameter ${\widehat{\mathrm{\θ}}}_i\left(0\right)=0.382*({\mathrm{\θ}}_{i\max }-{\mathrm{\θ}}_{i\min })+{\mathrm{\θ}}_{i\min }---\left(3\right)$ (3) set up full dose parameter estimation formula in the transient process stage, the full dose parameter estimation is represented with formula (4), (5) or (6): $P\left(k\right)=\frac{1}{\mathrm{\λ}}\{P(k-1)-P(k-1)\mathrm{\φ}\left(k\right){[{\mathrm{\φ}}^T\left(k\right)P(k-1)\mathrm{\φ}\left(k\right)+\mathrm{\λ}]}^{-1}{\mathrm{\φ}}^T\left(k\right)P(k-1)\}---\left(4\right)$ Wherein λ is a forgetting factor, generally gets 0.95～0.98.

Wherein

0＜λ ₁＜μ，0＜λ ₂＜4μ，0＜μ＜1，

The rectangular projection of ∏ (x) representative from x to Ω, the value of Ω is f ₁(x), f ₂(x), g ₀(x) scope.

d _n(k)+and d is the mould that characteristic model prediction equation value and practical object are exported the deviation that is allowed, this value generally can not considered.

(4) set up margin of error parameter estimation

In the control of position constant, as enter steady-state zone and error amount | e (k) |＞ε ₁The time (| e (k) |≤ε ₁The time, parameter stops to estimate), margin of error parameter estimation is used following formula (as having satisfied the control requirement with the full dose parameter estimation, this estimation of error formula can):

Error amount e (k)=y (k)-y _m(k) (7)

Y wherein _m(k) output valve for wishing, y (k) goes out value for actual elm. $e\left(k\right)=f_1\left(k\right)e(k-1)+f_2\left(k\right)e(k-2)+g_0\left(k\right)\mathrm{\Δu}(k-k_0-1)+g_1\left(k\right)\mathrm{\Δu}(k-k_0-2)$ $={\widetilde{\mathrm{\φ}}}^T\left(k\right)\mathrm{\θ}\left(k\right)+g_0\left(k\right)\mathrm{\Δu}(k-k_0-1)---\left(8\right)$ Wherein $\widetilde{\mathrm{\φ}}\left(k\right)={\left[\begin{array}{ccc}e(k-1)& e(k-2)& \mathrm{\Δu}(k-k_0-2)\end{array}\right]}^T,$

θ(k)＝[f ₁(k) f ₂(k) g ₁(k)] ^T

＝[θ ₁(k)?θ ₂(k)?θ ₃(k)] ^T。

The calculating of P (k) and formula (4) are same,

When system is minimum phase system, P (k) calculate with (4) with, (9) formula can substitute with (10) formula, promptly In (9), (10) formula, K _j(k) be the coefficient of feedback rate control, promptly $\mathrm{\Δu}\left(k\right)=-\{K_1\hat{e}(k+k_0/k)+K_2\hat{e}(k+k_0-1/k)+K_3\mathrm{\Δu}(k-1)\}.---\left(11\right)$ As hysteresis step number k ₀=0 o'clock, then Δ u (k)=-{ K ₁E (k)+K ₂E (k-1)+K ₃Δ u (k-1) }.For the minimum phase object, as hysteresis step number k ₀＞0 o'clock, have $\mathrm{\Δu}\left(k\right)=-\{K_{1}e\left(\frac{k+k_0}{k}\right)+K_{2}e\left(\frac{k+k_0-1}{k}\right)\}.$

For the minimum phase object, as hysteresis step number k ₀=0, have

Δu(k)＝-{K ₁e(k)+K ₂e(k-1)}，

The concrete formula of its Δ u (k) is seen below formula (14).

Above-mentioned full dose parameter estimation formula and margin of error parameter estimation formula all are based on the foundation of least-squares parameter estimation method, can also set up corresponding estimation formulas according to gradient method or other parameter estimation method.3. golden section design of Intelligence Controller

The present invention is for pure hysteresis step number k ₀=0 controlled device design golden section intelligent controller best results.

This controller is made up of four parts,

1. keep/the tracking Control rule;

2. golden-section adaptive control is restrained;

3. logic integral control law;

4. logic differential control law.

(1) keeps/the tracking Control rule

For realizing constant or second order tracking Control, at first to desirable reference model that requires of input signal design, promptly

y _m(k+1)＝m ₁y _m(k)+m ₂y _m(k-1)+m ₃y _r(k-1) (12)

And constant or tracking Control u ₀(k) can utilize estimated parameter and reference model y _m(k) be designed to $u_0\left(k\right)=\frac{y_m(k+1)-{\hat{f}}_1\left(k\right)y_m\left(k\right)-{\hat{f}}_2\left(k\right)y_m(k-1)}{\hat{g}\left(k\right)}---\left(13\right)$

M wherein _i(i=1,2,3) are by the determined parameter of control requirement.

(2) golden-section adaptive control rule $u_1\left(k\right)=\mathrm{\Δu}\left(k\right)=-\{l_1\frac{{\hat{f}}_1\left(k\right)}{\hat{g}\left(k\right)}(y\left(k\right)-y_m\left(k\right))+l_2\frac{{\hat{f}}_2\left(k\right)}{\hat{g}\left(k\right)}(y(k-1)-y_m(k-1))\}---\left(14\right)$

L wherein ₁=0.382, l ₂=0.618 is the golden section coefficient, ${\hat{f}}_1\left(k\right),{\hat{f}}_2\left(k\right),\hat{g}\left(k\right)$ Be respectively actual parameter f ₁(k), f ₂(k), the estimated value of g (k).

In fact, the actual formula of golden-section adaptive control rule should be $u_1\left(k\right)=\mathrm{\Δu}\left(k\right)=-r\{l_1\frac{{\hat{f}}_1\left(k\right)}{\hat{g}\left(k\right)}(y\left(k\right)-y_m\left(k\right))+l_2\frac{{\hat{f}}_2\left(k\right)}{\hat{g}\left(k\right)}(y(k-1)-y_m(k-1))\}---\left(15\right)$ Wherein r is relevant with target gain D,

(3) logic integral control law

General its integral coefficient of conventional integrator all is chosen to be normal value.The logic integral device then is according to system running state and control task requirement, judges that by analysis the back changes integral coefficient or integrator energy stored automatically, and concrete form is Wherein, work as y _r(k) raise y (k) c＞1 when not catching up with rapidly; Generalized case c=1; Work as y _r(k) descend y (k) c＜1 when not catching up with rapidly.

Use above-mentioned logic integral, can overcome step preferably and disturb, make the transient process performance improvement, and can follow the tracks of the hope curve of various sudden changes.

(4) logic differential control law

The ultimate principle and the logic integral of logic differential are similar, also are to change its differential strategy according to system running state.Its general form is $u_d\left(k\right)=-k_d{\left|\stackrel{\·}{y}\left(k\right)\right|}^{e_d}\mathrm{sgn}\left(\stackrel{\·}{y}\left(k\right)\right)---\left(17\right)$ Wherein: $k_d=\mathrm{cf}\left(y(k,\stackrel{\·}{y}\left(k\right))\right),$ c＞0， $f(y\left(k\right),\stackrel{\·}{y}\left(k\right))>0,$ e _d＞0, c, e _dValue depend on the output waveform Feature Recognition or rule of thumb determine, as $k_d=c\sqrt{\left|y\left(k\right)\stackrel{\·}{y}\left(k\right)\right|},$ c＝10，e _d＝1。

Transient process or the unloading phase, such differential action can effectively be improved dynamic quality.

(5) overhead control amount

u(k)＝u ₀(k)+u _i(k)+u _i(k)+u _d(k) (18)

U (k)=fu (k)+(1-f) u (k) | u (k) | u _Max(19) wherein f is a filter factor, 0＜f＜1.4. knowledge base and knowledge are revised

The defined knowledge of this intelligence control system is divided into three kinds substantially:

(1) true descriptive knowledge;

(2) knowledge of systematicness;

(3) theory, method, mature various computing formula.

Knowledge representation and generation are the key point of Based Intelligent Control.Obviously, for knowledge representation and generation problem, above-mentioned first kind knowledge can onlinely produce, and second, third class knowledge generally can only be imported by the expert, generate and express in order to solve this two, three class knowledge, we have proposed new ideas (as shown in Figure 2) such as dictionary, Runtime Library, in order to forming knowledge base, to satisfy in the intelligence control system actual motion demand to knowledge.

As shown in Figure 2, total program fundamental block diagram of the present invention thes contents are as follows: at first, system is carried out initialization, determine according to the result of measurement data whether system continues operation or shutdown then.If continue operation, next to carry out the nominal processing to measurement data, then, judge whether to call the knowledge in the dictionary, the knowledge in the dictionary can be imported by man-machine interaction.After as required the knowledge in the dictionary being called in the operation knowledge base, utilize the knowledge in the operation knowledge base to carry out reasoning, the result of reasoning determines the particular content of parameter estimation.Had after the new parameter estimation result, just can carry out the design of controller, then the controlled quentity controlled variable that calculates has been exported to topworks and carried out by the design of Intelligence Controller principle.Said process repeats, thereby realizes the control purpose of expectation.

Claims (1)

1. golden-section intelligent control method based on the characteristics of objects model description is characterized in that can pass to down step realizes:

A. the feature of controlled device is at first analyzed in modeling, promptly whether belong to linear stational system and can be linearizing the nonlinear constant system; Analyze the control performance requirement that above-mentioned object satisfies position maintenance and Position Tracking Control then, the difference equation of deformation type and four characteristic models that characteristic parameter is formed when requiring to set up a second order with control according to above-mentioned characteristics of objects, determine that specifically four characteristic parameters are as follows:

(1) determines or estimates the minimum equivalent time constant T of controlled device _Min

(2) definite or the static gain D of estimation controlled device or the scope of estimation static gain D;

(3) select rational time in sampling period Δ t, $\frac{\mathrm{\Δt}}{T_{\min }}=\frac{1}{5}~\frac{1}{15};$

(4) measure pure retardation time of τ, determine or estimate the pure hysteresis step number k of controlled device ₀, general calculating $k_0=\frac{\mathrm{\τ}}{\mathrm{\Δt}}$ Round numbers;

B. characteristic model is carried out parameter estimation,

(1), determines parameter estimation initial value and parameter estimation scope according to overall coefficient self-adaptation control method principle;

(2) in the transient process stage, utilize the output full dose to set up full dose parameter estimation formula, the parameter of characteristic model is estimated, and estimated result rectangular projection is arrived in the parameter estimation scope of determining the front that full dose parameter estimation formula is as follows: $P\left(k\right)=\frac{1}{\mathrm{\λ}}\{P(k-1)-P(k-1)\mathrm{\φ}\left(k\right)[{\mathrm{\φ}}^T\left(k\right)P(k-1)\mathrm{\φ}\left(k\right)+\mathrm{\λ}{]}^{-1}{\mathrm{\φ}}^T\left(k\right)P(k-1)\}$

Wherein λ is a forgetting factor, generally gets 0.95～0.98

Wherein

0＜λ ₁＜μ，0＜λ ₂＜4μ，0＜μ＜1，

The rectangular projection of Π (x) representative from x to Ω, the value of Ω is f ₁(x), f ₂(x), g ₀(x) scope,

d _n(k)+and d is the mould that characteristic model prediction equation value and practical object are exported the deviation that is allowed, this value generally can not considered;

(3) in the control of position constant, as enter steady-state zone and output error when bigger than setting value, utilize output error to set up margin of error parameter estimation formula, parameter to characteristic model is estimated, and in the parameter estimation scope that estimated result rectangular projection is determined to the front, concrete margin of error parameter estimation formula is:

Error amount e (k)=y (k)-y _m(k)

Y wherein _m(k) output valve for wishing, y (k) is a real output value $e\left(k\right)=f_1\left(k\right)e(k-1)+f_2\left(k\right)e(k-2)+g_0\left(k\right)\mathrm{\Δu}(k-k_0-1)+g_1\left(k\right)\mathrm{\Δu}(k-k_0-2)$ $={\widetilde{\mathrm{\φ}}}^T\left(k\right)\mathrm{\θ}\left(k\right)+g_0\left(k\right)\mathrm{\Δu}(k-k_0-1)$ Wherein $\widetilde{\mathrm{\φ}}\left(k\right)={\left[\begin{array}{ccc}e(k-1)& e(k-2)& \mathrm{\Δu}(k-k_0-2)\end{array}\right]}^T,$

θ(k)＝[f ₁(k) f ₂(k) g ₁(k)] ^T

＝[θ ₁(k)?θ ₂(k)?θ ₃(k)] ^T

The calculating of P (k) is the same,

K _j(k) be the coefficient of feedback rate control, promptly $\mathrm{\Δu}\left(k\right)=-\{K_1\hat{e}(k+k_0/k)+K_2\hat{e}(k+k_0-1/k)+K_3\mathrm{\Δu}(k-1)\};$

C. carry out Design of intelligent controller, (1) design is kept and tracking control unit; (2) design golden section feedback controller; (3) design logic integration and logic derivative controller;

D. set up three class knowledge bases (dictionary, Runtime Library and buffer pool), carry out the The whole control system software design.