CN101763035A - Method for controlling radial basis function (RBF) neural network tuned proportion integration differentiation (PID) and fuzzy immunization - Google Patents

Abstract

The invention relates to a method for controlling radial basis function (RBF) neural network tuned proportion integration differentiation (PID) and fuzzy immunization. A main loop adopts RBF neural network tuned PID control, and inputs of an RBF neural network comprise an error e and an output measured value yout; kp, ki and kd are obtained by calculation; after the variables are input to the PID control for carrying out calculation, a variable u1 is output; a subloop adopts fuzzy immunization control, the fuzzy immunization control adopts an error e2 output by the PID control and the variation rate as inputs, and an output is a nonlinear function f(x), and the fuzzy immunization control is used for controlling the subloop by immune calculation; and the control is applied to a serial control system so that the system hardly has overshoot in the process of transition and is more stable.

Description

RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control method

Technical field

The present invention relates to a kind of intelligent control technology, particularly a kind of RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control method that is used for cascade control system.

Background technology

The conduct of PID (Proportion Integration Differentiation, proportional-integral-differential) controller is existing more than 50 year history of controller of practicability the earliest, remains most widely used industrial control unit (ICU) now.The PID controller is easily understood, and does not need condition precedents such as accurate system model in the use, thereby becomes the controller that is most widely used.But PID is non-linear in control, the time when change, coupling and parameter and the uncertain complex process of structure, the work effect is undesirable.The most important thing is, how to transfer parameter all useless if the PID controller can not be controlled complex process.

Cascade control system is by the work of major and minor two controller serial connections, shown in Fig. 1 tandem control structure block diagram.Wherein, W ₁(s) be inertia district transport function; W ₂(s) be leading district transport function; PI is the inner looping proportional-plus-integral controller; PID is the external loop proportional-integral derivative controller; d ₁For interference is measured in output; d ₂For controlled quentity controlled variable is disturbed; Rin is a set-point; Yout is an outputting measurement value.The output of master controller is as the set-point of submaster controller, and the output of submaster controller removes to handle the control practical object, to realize the definite value control to variable.The purpose of cascade control system is in order to stablize master variable better, make it to equal set-point, and master variable to be exactly the output of major loop, and major loop is a fixed set point control system thus.The output of subloop is secondary variable, and the set-point of subloop is the output of master controller, so in cascade control system, secondary variable is not that requirement is constant, but require to change with the output of master controller, and be a following control system therefore.The purpose of cascade control system is in order to stablize master variable accurately, master variable to be had relatively high expectations, and it is poor generally not allow to have a surplus, so master controller is generally selected the PI control law, when object lags behind when big, also can introduce the suitable differential action.Requirement to secondary variable in the cascade control system is not tight.In control procedure, secondary variable is that the output of constantly following master controller changes and changes, so submaster controller has generally adopted P proportional control rule just, introduce suitable integral action in case of necessity, and the differential action generally is unwanted.When load was near setting valve, the result was better.But when under other load conditions, control performance is just relatively poor, even will switch to manual adjustments from automatic control.

Summary of the invention

The present invention be directed to cascade control system PID non-linear in control, the time when change, coupling and parameter and the uncertain complex process of structure, the unfavorable problem of work effect, a kind of RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control method have been proposed, with RBF neural network, immunological regulation mechanism and the organic combination of conventional PID control system, have good control effect and higher robustness.

Technical scheme of the present invention is: a kind of RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control method, it is characterized in that, and comprise following concrete steps:

1) RBF neural network tuned proportion integration differentiation PID control:

A) the PID I/O of RBF neural network tuned proportion integration differentiation: the RBF neural network be input as error e and the outputting measurement value yout of system, the RBF neural network is calculated output k _p, k _i, k _d, k _p, k _i, k _dAfter value input PID controller calculates, output variable u ₁

B) set the RBF neural network parameter: input vector X; Radially base vector is H; The center vector C of network; The sound stage width vector of network is B; Weight vector is W; The learning rate of network is η; Factor of momentum is α; Δ k _p, Δ k _i, Δ k _dLearning rate η _p, η _i, η _d

C) set the neural network tuned proportion integration differentiation index $E\left(k\right)=\frac{1}{2}{\mathrm{error}}^2\left(k\right),$ The outputting measurement value in error (k) the expression k moment and the deviation between the input value;

D) determine three inputs of PID:

xc(1)＝error(k)-error(k-1)

xc(2)＝error(k)

xc(3)＝error(k)-2error(k-1)+error(k-2)；

k _p, k _i, k _dAdjustment adopt the gradient descent method:

${\mathrm{\Δk}}_p=-{\mathrm{\η}}_p\frac{\∂E}{\∂k_p}=-{\mathrm{\η}}_p\frac{\∂E}{\∂y}\frac{\∂y}{\∂u}\frac{\∂u}{\∂k_p}={\mathrm{\η}}_p\mathrm{error}\left(k\right)\frac{\∂y}{\∂u}\mathrm{xc}\left(1\right)$

${\mathrm{\Δk}}_i=-{\mathrm{\η}}_i\frac{\∂E}{\∂k_i}=-{\mathrm{\η}}_i\frac{\∂E}{\∂y}\frac{\∂y}{\∂u}\frac{\∂u}{\∂k_i}={\mathrm{\η}}_i\mathrm{error}\left(k\right)\frac{\∂y}{\∂u}\mathrm{xc}\left(2\right)$

${\mathrm{\Δk}}_d=-{\mathrm{\η}}_d\frac{\∂E}{\∂k_d}=-{\mathrm{\η}}_d\frac{\∂E}{\∂y}\frac{\∂y}{\∂u}\frac{\∂u}{\∂k_d}={\mathrm{\η}}_d\mathrm{error}\left(k\right)\frac{\∂y}{\∂u}\mathrm{xc}\left(3\right)$

η wherein _p, η _i, η _dBe respectively Δ k _p, Δ k _i, Δ k _dLearning rate,

k _p＝k _p-1+Δk _p

k _i＝k _i-1+Δk _i

k _d＝k _d-1+Δk _d

E) computing controller output variable u ₁, u (k)=u (k-1)+k _pXc (1)+k _iXc (2)+k _dXc (3);

2) fuzzy immunization control:

1) I/O of fuzzy immunization control: the input variable of fuzzy immunization control is the error e of inner looping RBF neural network tuned proportion integration differentiation PID control output ₂And rate of change de ₂/ dt; Fuzzy immunization control output variable is f (x); Control system always is output as u ₂

2) membership function of design fuzzy immunization control input and output;

3) sum up the fuzzy immunization control law:, sum up the fuzzy immunization control law according to T cell feedback regulation principle;

4) stimulation that is subjected to according to cell comes analogy control law wherein, tries to achieve k _u=K (1-τ f (x)), wherein K is a controls reaction speed, and τ is the control stabilization effect, and f (x) is a selected nonlinear function;

5) according to k _u, controlled system always exports u ₂, u ₂=k _uE ₂

Beneficial effect of the present invention is: RBF neural network tuned proportion integration differentiation PID of the present invention and fuzzy immunization control method, apply in the cascade control system, and the system that makes does not almost have overshoot in transient process, and system is more stable.

Description of drawings

Fig. 1 is a tandem control structure block diagram;

Fig. 2 is RBF neural network structure figure;

Fig. 3 is RBF neural network tuned proportion integration differentiation PID of the present invention and fuzzy immunization control block diagram;

Fig. 4 is the membership function figure of deviation of the present invention and deviation variation rate;

Fig. 5 exports the membership function figure of f (x) for the present invention;

Fig. 6 is a unit-step response curve map of the present invention;

Fig. 7 is a system interference response curve of the present invention.

Embodiment

Along with the development of Intelligent Control Theory such as expert system, fuzzy control, neural network, formed the Intelligent PID Control strategy that multiple Based Intelligent Control combines with PID control.Because neural network has self-organization, self study, adaptive ability, become most important a kind of mode in the Based Intelligent Control based on the control of neural network.RBF (Radial Basis Function, radial basis function, hereinafter to be referred as RBF) neural network has the ability that can approach any Nonlinear Mapping, and network structure is simple, the connection weights of its output are linear with output, can adopt the linear optimization algorithm that guarantees global optimum, thereby become the focus of research.

The RBF neural network is three layers of feedforward network with single hidden layer.Because it has simulated the local neural network structure of adjusting, covering acceptance domain mutually in the human brain.Therefore, the RBF network is that network is approached in a kind of part, and it can approach any continuous function with arbitrary accuracy.

The RBF network is a kind of three layers of feedforward network, is non-linear by the mapping that is input to output, and the hidden layer space is linear to the mapping of output region, thereby has accelerated pace of learning greatly and avoided the local minimum problem.The RBF network structure as shown in Figure 2.

In the RBF network structure, X=[x ₁, x ₂, Λ x _n] ^T, be the input vector of network.If the radially base vector H=[h of RBF network ₁, h ₂, Λ h _jΛ h _m] ^T, h wherein _jBe the gaussian basis function

$h_j=\exp (-\frac{{\left|\right|X-C_j\left|\right|}^2}{2b_j^2}),j=\mathrm{1,2},\mathrm{\Λm}---\left(1\right)$

The center vector of j node of network is C _j=[c _J1, c _J2Λ c _JiΛ c _Jn] ^T, wherein, i=1,2, Λ n

If the sound stage width vector of network is: B=[b ₁, b ₂Λ Λ b _m] ^T, b _jBe the sound stage width degree parameter of node j, and be number greater than zero.The weight vector of network is W=[w ₁, w ₂Λ w _jΛ w _m] ^T(2)

Identification network is output as: y _m(k)=w ₁h ₁+ w ₂h ₂+ Λ w _mh _m(3)

The performance index function of identifier is: ${\mathrm{<figure-callout\; id="1"\; label="J"\; filenames="H2009101987170E00014.png,H2009101987170E00021.png"\; state="\{\{state\}\}">J</figure-callout>}}_1=\frac{1}{2}{(\mathrm{yout}\left(k\right)-y_m\left(k\right))}^2---\left(4\right)$

According to the gradient descent method, the iterative algorithm of output power, node center and node sound stage width parameter is as follows:

w _j(k)＝w _j(k-1)+η(yout(k)-y _m(k))h _j+α(w _j(k-1)-w _j(k-2)) (5)

$\mathrm{\&Delta;}{b}_j=(\mathrm{yout}\left(k\right)-y_m\left(k\right))w_j{h}_j\frac{{\left|\right|X-C_j\left|\right|}^2}{b_j^3}---\left(6\right)$

b _j(k)＝b _j(k-1)+ηΔb _j+α(b _j(k-1)-b _j(k-2)) (7)

${\mathrm{\&Delta;c}}_{\mathrm{ji}}=(\mathrm{yout}\left(k\right)-y_m\left(k\right))w_j\frac{x_j-c_{\mathrm{ji}}}{b_j^2}---\left(8\right)$

c _ji(k)＝c _ji(k-1)+ηΔc _ji+α(c _ji(k-1)-c _ji(k-2)) (9)

In the formula, η is a learning rate, and α is a factor of momentum.

Jacobian battle array (being the sensitivity information of the output of object to the control input) algorithm is

$\frac{\&PartialD;y\left(k\right)}{\&PartialD;u\left(k\right)}\&ap;\frac{\&PartialD;y_m\left(k\right)}{\&PartialD;u\left(k\right)}=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}{w}_j{h}_j\frac{c_{\mathrm{ji}}-x_1}{b_j^2}---\left(10\right)$

In the formula, x ₁=u (k)

RBF network PID three of the PID that adjust are input as:

xc(1)＝error(k)-error(k-1) (11)

xc(2)＝error(k) (12)

xc(3)＝error(k)-2error(k-1)+error(k-2) (13)

In the formula, the outputting measurement value in error (k) the expression k moment and the deviation between the input value.

Control algolithm is:

u(k)＝u(k-1)+k _pxc(1)+k _ixc(2)+k _dxc(3) (14)

The neural network tuned proportion integration differentiation index is:

$E\left(k\right)=\frac{1}{2}{\mathrm{error}}^2\left(k\right)---\left(15\right)$

k _p, k _i, k _dAdjustment adopt the gradient descent method:

${\mathrm{\&Delta;k}}_p=-{\mathrm{\&eta;}}_p\frac{\&PartialD;E}{\&PartialD;k_p}=-{\mathrm{\&eta;}}_p\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_p}={\mathrm{\&eta;}}_p\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}\left(1\right)---\left(16\right)$

${\mathrm{\&Delta;k}}_i=-{\mathrm{\&eta;}}_i\frac{\&PartialD;E}{\&PartialD;k_i}=-{\mathrm{\&eta;}}_i\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_i}={\mathrm{\&eta;}}_i\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}\left(2\right)---\left(17\right)$

${\mathrm{\&Delta;k}}_d=-{\mathrm{\&eta;}}_d\frac{\&PartialD;E}{\&PartialD;k_d}=-{\mathrm{\&eta;}}_d\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_d}={\mathrm{\&eta;}}_d\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}\left(3\right)---\left(18\right)$

In the formula,

Be the Jacobian information of controlled device, can get by the identification of neural network; η _p, η _i, η _dBe respectively Δ k _p, Δ k _i, Δ k _dLearning rate.

Therefore,

k _p＝k _p-1+Δk _p (19)

k _i＝k _i-1+Δk _i (20)

k _d＝k _d-1+Δk _d (21)

Artificial immune system is a kind of intelligent method of natural imitation immunologic function, by biosome to external world the mechanism and the noise of defence naturally of material restrain oneself, evolutionary learning technology such as teacherless learning, self-organization, provide new method for solving various engineering problems.The fuzzy immunization controller is to use for reference the immune mechanism of biosystem and a kind of gamma controller of designing, and it has good control effect and higher robustness.

Immune system is the specific character physiological reaction of biosome.Biological immune system can produce corresponding antibody and resist for the antigen of external infringement, after antigen and the antibodies, can produce a series of reaction, by phagocytosis or produce specific enzymes be used for damaging antigen.Biological immune system is made up of lymphocyte and antibody molecule, and lymphocyte (is respectively auxiliary cell T by the T cell that thymus gland produces again _HWith inhibition cell T _S) and the B cell that produces of marrow form.Invade body and after peripheral cell digestion, information is passed to the T cell when antigen, promptly pass to T _HCell and T _SCell stimulates the B cell then.The B cell produces antibody to eliminate antigen.When antigen more for a long time, the T in the body _HCell is also more, and T _SCell is less, thereby can produce more B cell.Along with the minimizing of antigen, the T in the body _SCytosis, it has suppressed T _HThe generation of cell, then the B cell is also along with minimizing.Through after a while at interval after, immune feedback system just tends to balance.Suppressing the mutual cooperation between mechanism and the primary feedback mechanism, is by Immune Feedback Mechanism the rapid reaction and the stable immune system of antigen to be finished.

Though immune system is very complicated, the adaptive ability of its antigen is fairly obvious.These intelligent behaviors of bioinformation system are for science and engineering field provide various theoretical reference and technical method.Based on above-mentioned T cell feedback regulation principle, propose the fuzzy immunization controller: the antigen quantity that defines k generation is ε (k), by the T of antigenic stimulus _HCell is output as T _H(k), T _SCell is T to the influence of B cell _S(k), then total stimulation of B cell acceptance is: S (k)=T _H(k)-T _S(k) (22)

In the formula, T _H(k)=k ₁ε (k), T _S(k)=k ₂F (Δ S (k)) ε (k), k ₁, k ₂Be coefficient.

The amount of supposing killer T cell is provided by B cellular activity differential, then the u of killer T cell _Kill(k) amount is: u _Kill(k)=k ₁ε (k)-k ₂F (Δ u _Kill(k)) ε (k)=K (1-τ f (Δ u _Kill(k))) ε (k)=k _uε (k) (23) formula (23) i.e. immunity feedback rule.

If f is (Δ u _Kill(k)) being f (x), is a selected nonlinear function.If be defined as:

f(x)＝1-exp(-x/a) (24)

A＞0 wherein is for changing the parameter of functional form.Mild more when the curve of the big more f of a (x), to all x, f (x) ∈ [0,1] is arranged all.

The stimulation that is subjected to according to cell comes analogy control law wherein, obtains:

k _u＝K(1-τ·f(x)) (25)

Wherein K is a controls reaction speed, and τ is the control stabilization effect, and f () is a selected nonlinear function, is used for representing that cell suppresses the size of ability.

The deviation e that is input as inner looping of fuzzy immunization controller ₂With deviation variation rate de ₂/ dt (is designated as ec ₂), be output as the value of nonlinear function f (x).

The parameter adjustment rule of fuzzy immunization control is determined regulation rule according to the principle that " it is intense big that cell is accepted thorn, and it is more little then to suppress ability " reaches " it is intense little that cell is accepted thorn, and it is big more then to suppress ability ".

Control system always is output as u ₂=k _uE ₂(26)

At the designing requirement in the major and minor loop of cascade control system,, a kind of RBF adjust PID and fuzzy immunization control system have been designed with RBF neural network, immunological regulation mechanism and the organic combination of conventional PID control system.RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control structure as shown in Figure 3, major loop adopts the PID control of RBF neural network tuned proportion integration differentiation, the RBF neural network be input as error e and outputting measurement value yout, by calculating k _p, k _i, k _d, input to the PID controller and calculate; Subloop adopts the fuzzy immunization controller, and the fuzzy immunization controller is with error e ₂And rate of change is output as a nonlinear function f (x) as input, calculates through immunity then to be used to control subloop.

Design of Controller comprises two parts: major loop controlling Design and subloop controlling Design.

Major loop RBF neural network tuned proportion integration differentiation PID design of Controller step is as follows:

1) the RBF PID controller I/O of adjusting: the RBF neural network be input as error e and outputting measurement value yout, by calculating k _p, k _i, k _d, inputing to after the PID controller calculates, output variable is u ₁

2) set the RBF neural network parameter: input vector X; Radially base vector is H; The center vector C of network; The sound stage width vector of network is B; Weight vector is W; The learning rate of network is η; Factor of momentum is α; Δ k _p, Δ k _i, Δ k _dLearning rate η _p, η _i, η _d

3) set neural network tuned proportion integration differentiation index E;

4) determine three input: xc of PID (1), xc (2), xc (3); k _p, k _i, k _dAdjustment adopt the gradient descent method;

5) computing controller output variable u ₁

Subloop fuzzy immunization design of Controller step is as follows:

1) I/O of fuzzy immunization controller: the input variable of fuzzy immunization controller is the error e of inner looping ₂And rate of change; Fuzzy immunization controller output variable is f (x); Control system always is output as u ₂

2) membership function of design fuzzy immunization controller input and output;

3) sum up the fuzzy immunization control law:, sum up the fuzzy immunization control law according to T cell feedback regulation principle;

4) stimulation that is subjected to according to cell comes analogy control law wherein, tries to achieve k _u

5) according to k _i, controlled system always exports u ₂, u ₂=k _uE ₂

In the boiler steam temperature automatic control system, superheat steam temperature is one of important indicator of boiler operatiopn quality, superheat steam temperature too high or too low security and economy that all can remarkable influence power plant.Superheat steam temperature is too high, may cause the high-pressure section metal of superheater, jet chimney and steam turbine to damage, thereby the upper limit of overheating steam temperature is not generally answered 5 ℃ of overrates.Superheat steam temperature is low excessively, can reduce the thermal efficiency of full factory again and influence the safety and economic operation of steam turbine, thereby the lower limit of overheating steam temperature generally is not less than 10 ℃ of ratings.The ratings of overheating steam temperature is usually more than 500 ℃, and for example high-pressure boiler is generally 540 ℃, and overheating steam temperature is remained in 540 ℃ the scope.Main cascade control system and the leading differential control system of adopting realizes in power plant.Adopt conventional PID cascade control method to be difficult to obtain satisfied control effect.Have the object of big inertia for this class of boiler overheating steam temperature, adopt conventional PID regulator, near the working point among a small circle in because its dynamic perfromance is similar to linearity, might control better; But, just need to revise pid parameter immediately, otherwise will not reach the control requirement when changing set-point on a large scale or being subjected to external environment (comprising operating mode) too during big disturbance.

At present, most thermal power plant's overheating steam temperature employings tandem controlling schemes shown in Figure 1.With certain power plant's overheating steam temperature is example, and its leading district transport function is: $W_2\left(s\right)=\frac{8}{{(1+15s)}^2},$ Its inertia district transport function is: $W_1\left(s\right)=\frac{1.125}{{(1+25s)}^3}.$

Major loop RBF Tuning PID Controller device design procedure is as follows:

1) the adjust input variable of PID of the RBF PID controller I/O of adjusting: RBF is error e and outputting measurement value yout; The controller output variable is u ₁

2) set the RBF neural network parameter: input vector X=[0,0,0] ^TRadially base vector is H=[0,0,0,0,0,0] ^TThe center vector of network $C=30*\left[\begin{array}{cccccc}1& 1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1& 1\\ 1& 1& 1& 1& 1& 1\end{array}\right];$ The sound stage width vector of network is B=40*[1,1,1,1,1,1] ^TWeight vector is W=10*[1,1,1,1,1,1] ^TThe learning rate of network is η=0.25; Factor of momentum is α=0.05; Learning rate η _p=0.2; η _i=0.2; η _d=0.2;

3) set neural network tuned proportion integration differentiation index E, $E\left(k\right)=\frac{1}{2}{\mathrm{error}}^2\left(k\right);$

4) determine three input: xc=[0 of PID, 0,0] ^Tk _p, k _i, k _dAdjustment adopt the gradient descent method;

5) computing controller output variable u ₁, u ₁(k)=u (k-1)+k _pXc (1)+k _iXc (2)+k _dXc (3).

Subloop fuzzy immunization design of Controller step is as follows:

1) I/O of fuzzy immunization controller: the input variable of fuzzy immunization controller is the error e of inner looping ₂And rate of change; Fuzzy immunization controller output variable is f (x); Control system always is output as u ₂

2) membership function of design fuzzy immunization controller input and output.Deviation e ₂Rate of change ec with deviation ₂Domain be [1,1], the domain of output is [1,1].The fuzzy set of each input variable is that { fuzzy set of output variable is { N, Z, P} for N, P}.Deviation, the membership function of deviation variation rate is as shown in Figure 4; The membership function of output f (x) as shown in Figure 5.

3) sum up the fuzzy immunization control law: according to T cell feedback regulation principle, sum up the fuzzy immunization control law,

If?e ₂?and?ec ₂?are?big，then?f(·)is?small；

If?e ₂?is?big?and?ec ₂?is?small，then?f(·)is?zero；

If?e ₂?is?small?and?ec ₂?is?big，then?f(·)is?zero；

If?e ₂?and?ec ₂?are?small，then?f(·)is?big.

4) stimulation that is subjected to according to cell comes analogy control law wherein, tries to achieve k _u=K (1-τ f (x)), wherein, controls reaction speed K=7.55; Control stabilization effect τ=0.1.；

5) according to k _u, controlled system always exports u ₂=k _uE ₂

Emulation by conventional PID controllers, RBF are adjusted PID and fuzzy immunization controller obtains following control curve as shown in Figure 6.

As shown in Figure 6, when being input as unit step, be 300s the settling time of traditional PI D, and 8% overshoot is arranged; And RBF adjusts and only is PID fuzzy immunization control method settling time 200s, does not almost have overshoot, and system is more stable.

Fig. 7 is conventional PID controllers, RBF PID and the fuzzy immunization controller family curve after system adds the unit step disturbance of adjusting.As shown in Figure 7, the overshoot of traditional PID control is 13%, and be 320s settling time; And the overshoot of new algorithm is 8%, and be 250s settling time.

Claims (1)

1. RBF neural network tuned proportion integration differentiation PID and fuzzy immunization control method is characterized in that, comprise following concrete steps:

1) RBF neural network tuned proportion integration differentiation PID control:

A) the PID I/O of RBF neural network tuned proportion integration differentiation: the RBF neural network be input as error e and the outputting measurement value yout of system, the RBF neural network is calculated output k _p, k _i, k _d, k _p, k _i, k _dAfter value input PID controller calculates, output variable u ₁

B) set the RBF neural network parameter: input vector X; Radially base vector is H; The center vector C of network; The sound stage width vector of network is B; Weight vector is W; The learning rate of network is η; Factor of momentum is α; Δ k _p, Δ k _i, Δ k _dLearning rate η _p, η _i, η _d

C) set the neural network tuned proportion integration differentiation index $E\left(k\right)=\frac{1}{2}{\mathrm{error}}^2\left(k\right),$ The outputting measurement value in error (k) the expression k moment and the deviation between the input value;

D) determine three inputs of PID:

xc(1)＝error(k)-error(k-1)

xc(2)＝error(k)

xc(3)＝error(k)-2error(k-1)+error(k-2)；

k _p, k _i, k _dAdjustment adopt the gradient descent method:

$\mathrm{\&Delta;}{k}_p=-{\mathrm{\&eta;}}_p\frac{\&PartialD;E}{\&PartialD;k_p}=-{\mathrm{\&eta;}}_p\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_p}={\mathrm{\&eta;}}_p\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}---\left(1\right)$

$\mathrm{\&Delta;}{k}_i=-{\mathrm{\&eta;}}_i\frac{\&PartialD;E}{\&PartialD;k_i}=-{\mathrm{\&eta;}}_i\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_i}={\mathrm{\&eta;}}_i\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}---\left(2\right)$

$\mathrm{\&Delta;}{k}_d=-{\mathrm{\&eta;}}_d\frac{\&PartialD;E}{\&PartialD;k_d}=-{\mathrm{\&eta;}}_d\frac{\&PartialD;E}{\&PartialD;y}\frac{\&PartialD;y}{\&PartialD;u}\frac{\&PartialD;u}{\&PartialD;k_d}={\mathrm{\&eta;}}_d\mathrm{error}\left(k\right)\frac{\&PartialD;y}{\&PartialD;u}\mathrm{xc}---\left(3\right)$

η wherein _p, η _i, η _dBe respectively Δ k _p, Δ k _i, Δ k _dLearning rate,

k _p＝k _p-1+Δk _p

k _i＝k _i-1+Δk _i

k _d＝k _d-1+Δk _d

E) computing controller output variable u ₁, u (k)=u (k-1)+k _pXc (1)+k _iXc (2)+k _dXc (3);

2) fuzzy immunization control:

1) I/O of fuzzy immunization control: the input variable of fuzzy immunization control is the error e of inner looping RBF neural network tuned proportion integration differentiation PID control output ₂And rate of change de ₂/ dt; Fuzzy immunization control output variable is f (x); Control system always is output as u ₂

2) membership function of design fuzzy immunization control input and output;

3) sum up the fuzzy immunization control law:, sum up the fuzzy immunization control law according to T cell feedback regulation principle;

4) stimulation that is subjected to according to cell comes analogy control law wherein, tries to achieve k _u=K (1-τ f (x)), wherein K is a controls reaction speed, and τ is the control stabilization effect, and f (x) is a selected nonlinear function;

5) according to k _u, controlled system always exports u ₂, u ₂=k _uE ₂

Images