CN103324093A - A multi-model adaptive control system and its control method - Google Patents

Abstract

The invention discloses a multi-model adaptive control system, which aims at the control of a linear bounded nonlinear discrete time system and comprises a linear robust adaptive controller, a nonlinear neural network adaptive controller and a switching mechanism, wherein the switching mechanism is respectively connected with two adaptive controllers and the input end of a controlled object, and the two adaptive controllers are connected with the output end of the controlled object; the linear robust adaptive controller comprises a linear robust adaptive model and a linear robust controller, and the nonlinear neural network adaptive controller comprises a nonlinear neural network adaptive model and a nonlinear neural network controller; the input of the controlled object is selectively generated between the two controllers by the switching structure, a closed-loop negative feedback is arranged between the output of the controlled object and the two adaptive controllers, and the controlled object and the model outputs of the two adaptive controllers are in a subtraction relation, so that a model error is calculated and used for adjusting the parameters of the model and the weight of the neural network.

Description

A kind of multi-model Adaptive Control system and control method thereof

Technical field

The present invention relates to the adaptive control field, relate in particular to a kind of multi-model Adaptive Control system and control method thereof.

Background technology

Existing most of real system all is nonlinear system, usually has model structure uncertain, unknown parameters, and parameter changes the characteristics such as frequent; Adaptive control can be carried out on-line identification to nonlinear system, and the systematic parameter of utilizing identification to obtain is come CONTROLLER DESIGN; This method can be widely used in the control of nonlinear system, yet when the uncertainty of system was very large, such as the change of operating conditions, the situation of the bad or loses stability of transient performance may appear in the self-adaptation control method of traditional single identifier.

The proposition of multi-model Adaptive Control method, solve this problem by comprising following two parts: 1) multi-controller, formed by a plurality of identifiers and candidate's controller, a plurality of identifiers can cover whole parameter space; 2) switching mechanism produces control inputs by selecting candidate's controller; At present the research of multi-model automatic correction controling method mainly concentrated on how to optimize the multi-model collection, reduce the Models Sets scale, reach the purpose that improves control response speed; And how to design switching mechanism, so that the multi-model self-aligning control system has better transient performance and robust performance.

The multi-model automatic correction controling method has certain requirement to non-linear controlled device, thereby satisfy stability and the robustness of nonlinear control system, this has had a strong impact on the method widespread use in practice, relax achievement aspect the assumed condition of controlled device less; At first, nonlinear system is carried out linearization, system is divided into linear segment and nonlinear high-order function item, requiring nonlinear terms is bounded, and nonlinear system is minimum phase system; By the introducing of k difference operator, nonlinear terms bounded is loosened to the rate of rise bounded of nonlinear terms; On the basis of nonlinear terms rate of rise bounded, the generalized minimum-variance adaptive controller is introduced into, and above-mentioned conclusion is generalized to non-minimum phase system; But these conditions are still stricter for the requirement of nonlinear system, need to further relax.

Therefore, we must propose a kind of multi-model neural network self-adaptation control method that relates to a class Bounded Linear non-linear complax control system, to solve in the prior art impact on the restrictive condition of nonlinear system etc.

Summary of the invention

In order to overcome the defective of prior art, the present invention aims to provide a kind of restrictive condition that can relax nonlinear system, and enlarges a kind of multi-model Adaptive Control system and the control method thereof of the usable range of multi-model Adaptive Control method.

To achieve these goals, the invention provides a kind of multi-model Adaptive Control system, control for the nonlinear discrete time system of a class Bounded Linear, comprise two Indirect adaptive control devices and switching mechanism, described switching mechanism connects respectively the input end of described two Indirect adaptive control devices and a controlled device, and described two Indirect adaptive control devices connect the output terminal of described controlled device; Be provided with a close loop negative feedback between the output terminal of described controlled device and described two the Indirect adaptive control devices, and be set to subtract each other relation between the output of the model of described controlled device and described two Indirect adaptive control devices; The input of described controlled device is selected to produce between described two Indirect adaptive control devices by described switching mechanism, thus the computation model error, for the weights of parameter and the neural network of adjusting described two models.

Preferably, described two Indirect adaptive control devices comprise linear Robust adaptive controller and nonlinear neural network adaptive controller.

Preferably, described linear Robust adaptive controller comprises linear robust adaptive model and linear robust controller; Described linear robust adaptive model is by the projection identification algorithm, guarantees that the Identification Errors of described linear robust adaptive model is bounded also when the restrictive condition of the nonlinear terms of described multi-model Adaptive Control system is loosened to Bounded Linear.

Preferably, described linear robust controller is linear pole-placement and adaptive control device, by described linear pole-placement and adaptive control device with the Assignment of Closed-Loop Poles of described multi-model Adaptive Control system to the expectation position, thereby obtain to process the ability of the control problem of non-minimum phase system and open-loop unstable system.

Preferably, described nonlinear neural network adaptive controller comprises nonlinear neural network adaptive model and nonlinear neural network controller, and described nonlinear neural network adaptive model is comprised of linear segment and non-linear partial, the coefficient of described linear segment upgrades as auto-adaptive parameter, and described non-linear partial is made of neural network; Described nonlinear neural network adaptive model is by the online weights of adjusting neural network, thereby acquisition is to the estimation output of described controlled device.

Preferably, described nonlinear neural network controller is the pole-placement and adaptive control device with nonlinear terms, improves the control accuracy of described multi-model Adaptive Control system by described pole-placement and adaptive control device with nonlinear terms.

Preferably, described switching mechanism is provided with the performance index module, and described performance index module comprises cumulative errors part and model error part, and described cumulative errors is partly for the frequent switching that prevents described multi-model Adaptive Control system; Described switching mechanism calculates the performance index of each controller by control the moment at each, thereby the less controller of selectivity index produces next control inputs constantly.

The invention allows for a kind of multi-model Adaptive Control method, comprise the steps:

S1: system initialization: the parameter of the linear robust adaptive model of random initializtion and nonlinear neural network adaptive model, and the weights of random initializtion neural network;

In the S2:k=0 moment, controlled device is output as 0; When k ≠ 0 moment, controlled device is output as the real output value of system, makes the poor departure e that obtains system with the setting value of system _c, the poor model error e that obtains is made in the output of the real output value of system and linear robust adaptive model ₁, the real output value of system and nonlinear neural network adaptive model are made the poor model error e that obtains ₂

S3: utilize respectively the parameter of the calculation of parameter controller of two models, with departure e _cAs the input of linear robust controller and nonlinear neural network controller, produce respectively controlled quentity controlled variable u by two controllers ₁And u ₂

S4: calculate the performance index value of linear robust controller and nonlinear neural network controller, and the input u that produces of the less controller of selectivity desired value _i, as the control inputs u of controlled device and linear robust adaptive model and nonlinear neural network adaptive model;

S5: upgrade respectively the parameter of linear robust adaptive model and nonlinear neural network adaptive model and the weights of neural network;

S6: forward step S2 to.

Preferably, described neural network is set to single hidden layer, and the number of the hidden neuron of described neural network is set to 6-10 usually.

Compared with prior art, beneficial effect of the present invention is as follows:

1, the present invention can find out by the multi-model Adaptive Control method, comprise a linear robust adaptive model in the linear Robust adaptive controller, this linearity robust adaptive model is by the projection identification algorithm of introducing with standardized correction, the restrictive condition of the nonlinear terms of controlled device is loosened to Bounded Linear, has widened greatly the scope of application of multi-model Adaptive Control system.

2, the present invention is by arranging the pole-placement and adaptive control device, the Assignment of Closed-Loop Poles of multi-model Adaptive Control system is arrived the position of expectation, embody requirement to the performance index of closed-loop system etc. by providing limit, use the pole-placement and adaptive control device can guarantee the stability of non-minimum phase system, thereby the restrictive condition of controlled device is loosened to non-minimum phase system from minimum phase system.

3, the present invention is by the design of switching mechanism, so that the controller of multi-model Adaptive Control system switches between linear Robust adaptive controller and nonlinear neural network adaptive controller, and the less controller of selectivity desired value is as the control inputs of current system; And the accumulation item that comprises an error in the performance index, so that system's output is comparatively level and smooth; And because linear Robust Adaptive Control system has stability, and the present invention adopts linear Robust adaptive controller and nonlinear neural network adaptive controller to switch, thereby so that multi-model Adaptive Control of the present invention system has stability.

Description of drawings

The multi-model Adaptive Control device that Fig. 1 designs for the present invention close the feedback control system block scheme;

Fig. 2 is the linear Robust adaptive controller structural drawing of the present invention;

Fig. 3 is nonlinear neural network adaptive controller structural drawing of the present invention;

Fig. 4 is the structural drawing of neural network of the present invention;

Fig. 5 is the switching flow figure of switching mechanism of the present invention;

Fig. 6 is the simulation experiment result figure.

The symbol tabulation:

M ₁-linear robust adaptive model, M ₂-nonlinear neural network adaptive model, C ₁-linear robust controller, C ₂-nonlinear neural network controller, 100-switching mechanism, 200-controlled device.

Embodiment:

Referring to the accompanying drawing that the embodiment of the invention is shown, hereinafter with more detailed description the present invention.Yet the present invention can be with realizations such as multi-form, specifications, and should not be construed as the restriction of the embodiment that is subjected in this proposition.On the contrary, it is abundant and complete open in order to reach proposing these embodiment, and makes more relevant those skilled in the art person understand scope of the present invention fully.In these accompanying drawings, for clearly visible, may zoom in or out relative size.

Referring now to Fig. 1 describes in detail according to multi-model Adaptive Control of the invention process system, this multi-model Adaptive Control system is multi-model Adaptive Control system and the control method thereof of a class Bounded Linear nonlinear system, relax the restrictive condition of nonlinear system, thereby enlarge the usable range of multi-model Adaptive Control method.This multi-model Adaptive Control system is comprised of a linear Robust adaptive controller, a neural network Indirect adaptive control device and a switching mechanism; Wherein this switching mechanism one end connects the input end of controlled device, and the other end connects respectively linear Robust adaptive controller and neural network Indirect adaptive control device; This linearity Robust adaptive controller is connected with controlled device with neural network Indirect adaptive control device and is connected, and be provided with a close loop negative feedback between two controllers and the controlled device, and be set to subtract each other relation between the output of the model of controlled device and two controllers, thereby calculate model error, and this model error is used for the parameter of adjustment model and the weights of neural network.

Wherein, linear Robust adaptive controller comprises linear robust adaptive model and linear robust controller, linear segment behind the corresponding multi-model Adaptive Control system linearization of linear robust adaptive model, it is a kind of linear autoregressive moving average input/output model, adaptive parameter is the coefficient of linear regression vector, adaptive law adopts is with standardized correction projection identification algorithm, utilize identified parameters design Pole Assignment Controller according to definite equivalence principle, by the projection identification algorithm, when assurance was loosened to Bounded Linear when the restrictive condition of the nonlinear terms of multi-model Adaptive Control system, the Identification Errors of linear robust adaptive model is bounded also.And, linear robust controller is linear pole-placement and adaptive control device, by the pole-placement and adaptive control device with the Assignment of Closed-Loop Poles of multi-model Adaptive Control system to the expectation position, thereby obtain to process the ability of the control problem of non-minimum phase system and open-loop unstable system.

And the nonlinear neural network adaptive controller comprises nonlinear neural network adaptive model and nonlinear neural network controller, and the nonlinear neural network adaptive model obtains the estimation output to controlled device by the online weights of adjusting neural network; The nonlinear neural network adaptive model is comprised of linear segment and non-linear partial, the coefficient of linear segment can upgrade in any way as auto-adaptive parameter, non-linear partial is made of neural network, and this neural network adopts the BP neural network, utilize the error back propagation method to train, and according to definite equivalence principle, utilize identified parameters and the neural network model that obtains, design non-linear pole-placement and adaptive control device; Improve the control accuracy of multi-model Adaptive Control system by non-linear pole-placement and adaptive control device.

In addition, switching mechanism is provided with the performance index module, and the performance index module comprises cumulative errors part and model error part, and cumulative errors is partly for the frequent switching that prevents the multi-model Adaptive Control system; Switching mechanism calculates the performance index of each controller by control the moment at each, thereby the less controller of selectivity index produces next control inputs constantly.

The invention allows for a kind of multi-model Adaptive Control method, comprise the steps:

S1: system initialization: the parameter of the linear robust adaptive model of random initializtion and nonlinear neural network adaptive model, and the weights of random initializtion neural network;

In the S2:k=0 moment, controlled device is output as 0; When k ≠ 0 moment, controlled device is output as the real output value of system, makes the poor departure e that obtains system with the setting value of system _c, the poor model error e that obtains is made in the output of the real output value of system and linear robust adaptive model ₁, the real output value of system and nonlinear neural network adaptive model are made the poor model error e that obtains ₂

S3: utilize respectively the parameter of the calculation of parameter controller of two models, with departure e _cAs the input of linear robust controller and nonlinear neural network controller, produce respectively controlled quentity controlled variable u by two controllers ₁And u ₂

S4: calculate the performance index value of linear robust controller and nonlinear neural network controller, and the input u that produces of the less controller of selectivity desired value _i, as the control inputs u of controlled device and linear robust adaptive model and nonlinear neural network adaptive model;

S5: upgrade respectively the parameter of linear robust adaptive model and nonlinear neural network adaptive model and the weights of neural network;

S6: forward step S2 to.

Wherein, neural network is set to single hidden layer, and the number of the hidden neuron of neural network is set to 6-10 usually.

Application examples

As shown in Figure 1, in the designed controller of multi-model Adaptive Control method of the present invention, formed by a linear Robust adaptive controller, a nonlinear neural network adaptive controller and a switching mechanism.Among the figure, r (k+n) is the tracking reference signal of multi-model Adaptive Control system, u _p(k) be the input of controlled device, y _p(k+n) be the output of controlled device; Linear Robust adaptive controller comprises linear robust adaptive model M ₁With linear pole-placement and adaptive control device C ₁, u ₁(k) be linear pole-placement and adaptive control device C ₁Output, y _p(k+n) be linear robust adaptive model M ₁Output; The nonlinear neural network adaptive controller comprises nonlinear neural network adaptive model M ₂With non-linear pole-placement and adaptive control device C ₂, u ₂(k) be non-linear pole-placement and adaptive control device C ₂Output, y ₂(k+n) be nonlinear neural network adaptive model M ₂Output.u _p(k) by switching mechanism at u ₁(k) and u ₂(k) select between to produce.

Constant nonlinear system during for following input/output format discrete:

$y_p(k+n)=-\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}{a}_j{y}_p(k+j)+\underset{j=0}{\overset{M}{\mathrm{\Σ}}}{b}_j{u}_p(k+j)+f\left(w\left(k\right)\right)$

$=\mathrm{\θw}\left(k\right)+f\left(w\left(k\right)\right)---\left(1\right)$

θ in the formula=[θ _a, θ _b], θ _a=[a _N-1, a _N-2..., a ₀] ^T, θ _b=[b _m, b _M-1..., b ₀] ^T, w (k)=[y _p(k+n-1) ... ,-y _p(k), u _p(k+m) ..., u _p(k)] ^T

Make α _n=[z ⁿ..., z, 1], $A(z,k)=z^n+{\mathrm{\θ}}_a^T{\mathrm{\α}}_{n-1}=z^n+a_{n-1}{z}^{n-1}+\·\·\·+a_0,$ $B(z,k)={\mathrm{\θ}}_b^T{\mathrm{\α}}_m=b_m{z}^m$ $+b_{m-1}{z}^{m-1}+\·\·\·+b_0,$ Then system can be expressed as:

A(z,k)y _p(k)=B(z,k)u _p(k)+f(w(k)) (2)

Filtering, Λ are carried out in system (1) both sides _pBe the Hurwitz polynomial expression of n for order, its root all exists The zone in, 0≤δ ₀＜1.

$\frac{z^n{y}_p\left(k\right)}{{\mathrm{\Λ}}_p}={\mathrm{\θ}}^T\frac{w\left(k\right)}{{\mathrm{\Λ}}_p}+\frac{f\left(w\left(k\right)\right)}{{\mathrm{\Λ}}_p}---\left(3\right)$

Order $\mathrm{\ζ}=\frac{z^n{y}_p\left(k\right)}{{\mathrm{\Λ}}_p},\mathrm{\φ}\left(k\right)=\frac{w\left(k\right)}{{\mathrm{\Λ}}_p},\mathrm{\η}\left(k\right)=\frac{f\left(w\left(k\right)\right)}{{\mathrm{\Λ}}_p},$ Then system (1) can be write as

ζ(k)=θ ^Tφ(k)+η(k) (4)

System (3) is provided following hypothesis:

Suppose 1: order n is known in system, and m≤n-1;

Suppose 2: suppose | η (k) |≤μ || φ (k) ||+γ, μ〉0, γ〉0;

Suppose 3: θ ∈ Ω, It is known compacting.

By linear robust Indirect adaptive control structural drawing shown in the accompanying drawing 2, this controller comprises that linear robust adaptive model and linear pole-placement and adaptive control device two parts form, and set up system linearity robust adaptive model M ₁:

${\widehat{\mathrm{\ζ}}}_1\left(k\right)={\mathrm{\θ}}_1^T(k-1)\mathrm{\φ}\left(k\right)---\left(5\right)$

θ in the formula ₁(k)=[b _1,0(k) ..., b _{1, m}(k), a _1,0(k) ..., a _{1, n-1}(k)] ^T, be model M ₁In k parameter constantly.

According to hypothesis 3, can obtain a N ₀0, so that have for all θ ∈ Ω

Model M ₁Parameter adopt following projection identification algorithm to upgrade, wherein

θ ₁(t) identification:

${\stackrel{\‾}{\mathrm{\θ}}}_1\left(k\right)={\widehat{\mathrm{\θ}}}_1(k-1)+{\mathrm{\Γ}}_1{\mathrm{\ϵ}}_1\mathrm{\φ}\left(k\right),{\mathrm{\θ}}_1\left(0\right)\∈\mathrm{\Ω}---\left(6\right)$

${\widehat{\mathrm{\θ}}}_1\left(k\right)=\left\{\begin{array}{cc}{\stackrel{\‾}{\mathrm{\θ}}}_1\left(k\right)& \mathrm{if}\left|{\stackrel{\‾}{\mathrm{\θ}}}_1\left(k\right)\right|\≤N_0\\ \frac{N_0}{\left|{\stackrel{\‾}{\mathrm{\θ}}}_1\left(k\right)\right|}{\stackrel{\‾}{\mathrm{\θ}}}_1\left(k\right)& \mathrm{otherwise}\end{array}\right.---\left(7\right)$

${\mathrm{\ϵ}}_1\left(k\right)=\frac{\mathrm{\ζ}\left(k\right)-{\widehat{\mathrm{\θ}}}_1^T(k-1)\mathrm{\φ}\left(k\right)}{m_s^2\left(k\right)}---\left(8\right)$

$m_s^2\left(k\right)={(\mathrm{\τ}+|\left|\mathrm{\φ}\left(k\right)\right|\left|\right)}^2+n_d\left(k\right)---\left(9\right)$

n _d(k+1)=δ ₀n _d(k)+|u _p(k)| ²+|y _p(k)| ² (10)

In the formula, n _d(0)=0,0＜Γ ₁＜2, ε ₁It is standardized model error.

Make θ ₁=[θ _1a, θ _1b], θ _1a=[a _{1, n-1}, a _{1, n-2}..., a _1,0] ^T, θ _1b=[b _{1, m}, b _{1, m-1}..., b _1,0] ^T,

${\hat{A}}_1(z,k)=z^n+{\mathrm{\θ}}_{1a}{\mathrm{\α}}_{n-1},$ ${\hat{B}}_1(z,k)={\mathrm{\θ}}_{1b}{\mathrm{\α}}_m.$

According to the linear pole-placement and adaptive control device of linear robust adaptive modelling C ₁:

${\hat{L}}_1(z,k)Q_m\left(z\right)u_1\left(k\right)=-{\hat{P}}_1(z,k)(y_p\left(k\right)-y_m\left(k\right))---\left(11\right)$

Determined by following formula

With

${\hat{A}}_1(z,k){\hat{L}}_1(z,k)Q_m\left(z\right)+{\hat{P}}_1(z,k){\hat{B}}_1(z,k)=A^{*}\left(z\right)---\left(12\right)$

In the formula ${\hat{L}}_1(z,k)=z^{n-1}+{{\stackrel{\‾}{l}}_1}^T{\mathrm{\α}}_{n-2},{\hat{P}}_1(z,k)=p_{10}{z}^{n+q-1}+{{\stackrel{\‾}{p}}_1}^T{\mathrm{\α}}_{n+q-2},$ A ^*(z) be the closed loop proper polynomial of Hurwitz.

By the structural drawing of nonlinear neural network adaptive controller shown in the accompanying drawing 3, this controller comprises nonlinear neural network adaptive model M ₂With non-linear pole-placement and adaptive control device C ₂Two parts.

Set up neural network model M ₂:

${\widehat{\mathrm{\ζ}}}_2\left(k\right)={\mathrm{\θ}}_2^T(k-1)\mathrm{\φ}\left(k\right)+\widehat{\mathrm{\η}}(W\left(k\right),k)---\left(13\right)$

The nonlinear function that is the bounded of Neural Networks Representation approaches, and W (k) is the weight coefficient of neural network, upgrades with the error back propagation method.

θ ₂(k) utilize the projection identification algorithm of revising to determine,

θ ₂(t) identification:

${\stackrel{\‾}{\mathrm{\θ}}}_2\left(k\right)={\widehat{\mathrm{\θ}}}_2(k-1)+{\mathrm{\Γ}}_2{\mathrm{\ϵ}}_2\mathrm{\φ}\left(k\right),{\widehat{\mathrm{\θ}}}_2\left(0\right)\∈\mathrm{\Θ}---\left(14\right)$

${\widehat{\mathrm{\θ}}}_2\left(k\right)=\left\{\begin{array}{cc}{\stackrel{\‾}{\mathrm{\θ}}}_2\left(k\right)& \mathrm{if}\left|{\stackrel{\‾}{\mathrm{\θ}}}_2\left(k\right)\right|\≤N_0\\ \frac{N_0}{\left|{\stackrel{\‾}{\mathrm{\θ}}}_2\left(k\right)\right|}{\stackrel{\‾}{\mathrm{\θ}}}_2\left(k\right)& \mathrm{otherwise}\end{array}\right.---\left(15\right)$

${\mathrm{\ϵ}}_2\left(k\right)=\frac{\mathrm{\ζ}\left(k\right)-{\widehat{\mathrm{\θ}}}_2^T(k-1)\mathrm{\φ}\left(k\right)-\widehat{\mathrm{\η}}(W\left(k\right),k)}{m_s^2\left(k\right)}---\left(16\right)$

$m_s^2\left(k\right)={(\mathrm{\τ}+|\left|\mathrm{\φ}\left(k\right)\right|\left|\right)}^2+n_d\left(k\right)---\left(17\right)$

n _d(k+1)=δ ₀n _d(k)+|u _p(k)| ²+|y _p(k)| ² (18)

In the formula,

n _d(0)=0,0＜Γ ₂＜2, ε ₂It is standardized model error.

Order

θ _2a=[a _2,n-1,a _2,n-2,…,a _2,0] ^Tθ _2b=[b _2,m,b _2,m-1,…,b _2,0] ^T

${\hat{B}}_2(z,k)={\mathrm{\θ}}_{2b}{\mathrm{\α}}_m.$

According to nonlinear neural network adaptive model M ₂, can obtain the non-linear pole-placement and adaptive control device C of system ₂:

${\hat{L}}_2(z,k)Q_m\left(z\right)u_2\left(k\right)=-{\hat{P}}_2(z,k)(y_p\left(k\right)-y_m\left(k\right))+F\left(z\right)\widehat{\mathrm{\η}}(W\left(k\right),k)---\left(19\right)$

F (z) is feedback gain, is determined by following formula

With

${\hat{A}}_2(z,k){\hat{L}}_2(z,k)Q_m\left(z\right)+{\hat{P}}_2(z,k){\hat{B}}_2(z,k)=A^{*}\left(z\right)---\left(20\right)$

In the formula ${\hat{L}}_2(z,k)=z^{n-1}+{{\stackrel{\‾}{l}}_2}^T{\mathrm{\α}}_{n-2},{\hat{P}}_2(z,k)=p_{20}{z}^{n+q-1}+{\stackrel{\‾}{p}}_2^T{\mathrm{\α}}_{n+q-2}.$

Structural drawing by neural network shown in the accompanying drawing 4, this neural network is for comprising these three layers neuronic BP neural network of input layer, hidden layer and output layer, each neuron between the levels connects entirely, does not connect between every layer of neuron, and input layer to the connection weight in middle layer is l _Ij, i=1,2 ..., n _a+ n _b-2, j=1,2 ..., p; Hidden layer to the connection weight of output layer is v _J1, j=1,2 ..., p; Each unit output threshold value of hidden layer is τ _j, j=1,2 ..., p; The output threshold value of output layer unit is γ ₁Parameter k=1,2 ..., m is input as w (k)=[y _p(k+n-1) ... ,-y _p(k), u _p(k+m) ..., u _p(k)] ^T

Each neuronic input s of hidden layer _jFor:

Use s _jCalculate each neuronic output b of hidden layer by transport function _jFor: b _j=g (s _j), j=1,2 ..., p. utilizes the output b of hidden layer _j, connection weight v _J1And threshold gamma ₁Calculate the neuronic output of output layer L _tFor:

Then calculate the neuronic response of output layer by transport function

For:

Utilize connection weight v _J1, error e ₂And the output b of hidden layer (k), _j, calculate each neuronic error d of hidden layer _j(t).

d _j(k)=[e ₂(k)v _j1]b _j(1-b _j) (21)

Utilize output error e ₂(k) with each neuronic output b of hidden layer _jRevise connection weight v _J1And threshold gamma ₁:

v _j1=v _j1+κe ₂(k)b _j

γ ₁=γ ₁+κe ₂(k)

j=1,2,L,p,0<κ<1

Utilize the error d of hidden neuron _j(k), the input w (k) of input layer=[y _p(k+n-1) ... ,-y _p(k), u _p(k+m) ..., u _p(k)] ^TRevise connection weight l _IjAnd threshold tau _j:

l _ij=l _ij+σd _j(k)w _i(k)

τ _j=τ _j+σd _j(k)

i=1,2,…,n _a+n _b-2,j=1,2,…,p,0<σ<1.

Shown in the process flow diagram of the switching mechanism of accompanying drawing 5, at first be respectively design performance index J of linear robust Indirect adaptive control and nonlinear neural network Indirect adaptive control ₁(k) and J ₂(k), its computing method are as follows:

$J_i\left(k\right)=\underset{l=1}{\overset{k}{\mathrm{\&Sigma;}}}\frac{{\mathrm{\&Gamma;}}_0}{2}{\mathrm{\&epsiv;}}_i^2\left(l\right)+c\underset{l=k-N+1}{\overset{k}{\mathrm{\&Sigma;}}}{e}_i^2\left(l\right),i=\mathrm{1,2}---\left(22\right)$

E in the formula _i(k)=ε _im _s, Γ ₀0, c 〉=0 is the constant of definition.

At each system time, calculate respectively J ₁And J ₂Value, the controller switching signal $\mathrm{\&sigma;}\left(k\right)=\left\{\begin{array}{cc}1& J_1\left(k\right)\&le;J_2\left(k\right)\\ 2& J_1\left(k\right)>J_2\left(k\right)\end{array}\right..$

Multi-model self tuning controller C is defined as:

u _p(k)=u _σ(k) (23)

Accompanying drawing 6 is multi-model Adaptive Control system experimentation result of the present invention, "-" multi-model Adaptive Control device of the present invention among Fig. 6 (a), and "--" is linear Robust adaptive controller, ". " is reference signal; From figure, can find out the output valve at different time of different controlling values; Ordinate is that switching construction switches selective value among Fig. 6 (b), and wherein 1 expression switching mechanism selects linear Robust adaptive controller as control output; 2 expression switching mechanisms select the nonlinear neural network adaptive controller as control output.

According to top discussion, the implementation On-line Control step of multi-model Adaptive Control method of the present invention is as follows:

S1: system initialization: the linear robust adaptive model M of random initializtion ₁With nonlinear neural network adaptive model M ₂Parameter and the weights of neural network, can determine according to priori;

In the S2:k=0 moment, system is output as zero, i.e. y (0)=0; When k ≠ 0 moment, provided the true output valve y of system by the controlled device of system _p(k), by model M ₁And M ₂Provide respectively the estimation output y of model ₁(k) and y ₂(k);

$y_1(k+1)={{\widehat{\mathrm{\&theta;}}}_1}^T\left(k\right)\mathrm{\&psi;}\left(k\right)$

$y_2(k+1)={{\mathrm{\&theta;}}_2}^T\left(k\right)w\left(k\right)+f_2(W\left(k\right),\stackrel{\&OverBar;}{w}\left(k\right))$

S3: the evaluated error of computation model is respectively e ₁(k)=y (k)-y ₁(k) and e ₂(k)=y (k)-y ₂(k);

S4: by the reference input r (k) of system and the true departure e that exports y (k) computing system of system _c(k);

S5: utilize model M ₁And M ₂Parameter come CONTROLLER DESIGN C ₁And C ₂Parameter, according to formula (5) (8), by the departure e of system _c(k) calculate respectively the output valve u of linear robust controller and nonlinear neural network controller ₁(k) and u ₂(k);

S6: the performance index J that is calculated each controller by the model evaluated error ₁(k) and J ₂(k);

S7: refer to the controller u that target value is less by switching mechanism formula (11) selectivity _i(k) as the control inputs u of controlled device _p(k);

S8: by model evaluated error e ₁(k) and e ₂(k), according to adaptive law (6) and (14) separately, upgrade respectively linear robust adaptive model M ₁With nonlinear neural network adaptive model M ₂Parameter and the weights of neural network;

S9: get back to step S2.

Indirect adaptive control device provided by the invention is not limited in linear Robust adaptive controller and the nonlinear neural network adaptive controller that the present embodiment proposes, can also comprise other nonlinear adaptive controllers, thereby realize multi-model Adaptive Control; And the neural network that the present invention proposes is not limited with the BP neural network that this proposes, and the number that the number of the hidden neuron of neural network also is not limited only to the present embodiment proposition is limited, can also be set to the neural network of other numbers of plies, thereby can train according to Back Propagation Algorithm, and the BP neural network that the present embodiment proposes only is the neural network model of a widespread use.

Obviously, those skilled in the art can carry out various changes and distortion to the present invention and not break away from the spirit and scope of the present invention.Like this, if these modifications of the present invention and distortion belong in the scope of claim of the present invention and equivalent technologies thereof, then the present invention also is intended to comprise these changes interior.

Claims (9)

1. A multi-model adaptive control system, aimed at a class of linearly bounded nonlinear discrete-time systems, is characterized in that it includes two indirect adaptive controllers and switching mechanisms, and the switching mechanisms are respectively connected to the indirect self-adaptive controllers. An adaptive controller and an input terminal of a controlled object, the indirect adaptive controller is connected to the output terminal of the controlled object; the output terminal of the controlled object is set between the two indirect adaptive controllers There is a closed-loop negative feedback, and the model output between the controlled object and the two indirect adaptive controllers is set to a subtractive relationship; the input of the controlled object is controlled by the switching mechanism between the two indirect adaptive controllers. Selection between adaptive controllers is generated. 2. The multi-model adaptive control system according to claim 1, wherein the two indirect adaptive controllers comprise a linear robust adaptive controller and a nonlinear neural network adaptive controller respectively. 3. The multi-model adaptive control system according to claim 2, wherein the linear robust adaptive controller comprises a linear robust adaptive model and a linear robust controller; the linear robust adaptive The model ensures that the identification error of the linear robust adaptive model is also bounded when the constraint condition of the nonlinear term of the multi-model adaptive control system is relaxed to be linearly bounded through the projection identification algorithm. 4. The multi-model adaptive control system according to claim 3, wherein the linear robust controller is a linear pole configuration adaptive controller, and the multi-model adaptive controller is configured by the linear pole configuration adaptive controller. The closed-loop poles of the model adaptive control system are configured to desired positions, so that the ability to deal with the control problems of non-minimum phase systems and open-loop unstable systems is obtained. 5. The multi-model adaptive control system according to claim 2, wherein the nonlinear neural network adaptive controller comprises a nonlinear neural network adaptive model and a nonlinear neural network controller, and the nonlinear neural network controller The linear neural network adaptive model is composed of a linear part and a nonlinear part, the coefficients of the linear part are updated as adaptive parameters, and the nonlinear part is composed of a neural network; the nonlinear neural network adaptive model is adjusted online The weight of the neural network, so as to obtain the estimated output of the controlled object. 6. multi-model adaptive control system according to claim 5, is characterized in that, described nonlinear neural network controller is a pole configuration adaptive controller with nonlinear term, through described with nonlinear term The pole configuration adaptive controller improves the control accuracy of the multi-model adaptive control system. 7. The multi-model adaptive control system according to claim 1, wherein the switching mechanism is provided with a performance index module, and the performance index module includes a cumulative error part and a model error part, and the cumulative error part It is used to prevent frequent switching of the multi-model adaptive control system; the switching mechanism calculates the performance index of each controller at each control moment, so as to select the controller with a smaller performance index to generate the control input at the next moment. 8. A multi-model adaptive control method, utilizing the multi-model adaptive controller as claimed in claim 1 to realize the control of the controlled object, is characterized in that, comprises the steps: S1: System initialization: Randomly initialize the parameters of the linear robust adaptive model and the nonlinear neural network adaptive model, and randomly initialize the weights of the neural network; S2: When k=0, the output of the controlled object is 0; when k≠0, the output of the controlled object is the actual output value of the system, and the control error e _c of the system is obtained by making a difference with the set value of the system, and the system The difference between the actual output value of the system and the output of the linear robust adaptive model is obtained to obtain the model error e ₁ , and the difference between the actual output value of the system and the nonlinear neural network adaptive model is obtained to obtain the model error e ₂ ; S3: Using the parameters of the two models to calculate the parameters of the controller, the control error e _c is used as the input of the linear robust controller and the nonlinear neural network controller, and the two controllers generate control quantities u ₁ and u ₂ respectively ; S4: Calculate the performance index value of the linear robust controller and the nonlinear neural network controller, and select the input u _i generated by the controller with the smaller performance index value as the controlled object and the linear robust adaptive model and the nonlinear neural network controller The control input u of the neural network adaptive model; S5: Respectively update the parameters of the linear robust adaptive model and the nonlinear neural network adaptive model and the weights of the neural network; S6: go to step S2. 9. The multi-model adaptive control method according to claim 8, wherein the neural network is set to a single hidden layer, and the number of neurons in the hidden layer of the neural network is set to 6-10.

Images