CN102998973B - Multi-model adaptive controller and control method of nonlinear system - Google Patents

Abstract

The invention discloses a multi-model adaptive controller and a control method for a nonlinear system. A nonlinear robust indirect adaptive controller and a nonlinear neural network indirect adaptive controller are adopted, and based on performance indicators at each sampling time Switch to the optimal controller for control. Compared with the traditional nonlinear multi-model adaptive controller and control method, the present invention relaxes the limit of the nonlinear term of the nonlinear system to zero-order close to bounded, which can effectively expand the adaptability of the multi-model adaptive controller .

Description

A multi-model adaptive controller and control method for nonlinear systems

technical field

The invention relates to various actual controlled objects in the production process of various industries, in particular to a class of multi-model self-adaptive controller of a class of zero-order bounded nonlinear complex control system and its control method.

Background technique

Due to the transformation of modern industry into technology-intensive, complex industrial system processes composed of a wide variety of subsystems and components with complex internal relationships have emerged. Such systems often have the characteristics of strong nonlinearity, fast time-varying, and uncertain models. Although the existing nonlinear processing methods have some applications, they are not perfect in theory, have high requirements on the controlled object, and are not general.

Due to the characteristics of a large number of unknown adaptive parameters, it is difficult for the methods in classical control theory and modern control theory to solve these characteristics. Adaptive control can deal with the control problems of a certain degree of uncertainty system and is widely used in the control of nonlinear systems. . Through system identification, adaptive control can automatically compensate for changes in model order, parameters, and input signals, and better control effects can be obtained. However, for the controlled object with fast time-varying and parameter jumps, the identification effect of the adaptive control system is not good, resulting in poor dynamic performance of the system and excessive transient errors. Multi-model adaptive control is developed on the basis of adaptive control, which can effectively reduce the transient error of the system. Therefore, a new model structure of the nonlinear system is proposed. This model structure is composed of a linear part and a nonlinear part. Control effect. However, the traditional multi-model adaptive control method requires the nonlinear term of the plant to be partially globally bounded, which limits the application of the multi-model adaptive control method in practical systems.

Contents of the invention

The present invention aims at the technical problems existing in the above-mentioned prior art, and provides a multi-model adaptive controller and a control method for a nonlinear system. Relax the constraints of nonlinear systems and expand the scope of use of multi-model adaptive control methods. This method not only relaxes the global bounded condition of the nonlinear term to zero-order nearly bounded, but also reduces the steady-state error of the multi-model adaptive control system and improves the control precision.

For achieving the above object, the technical scheme that the present invention takes is:

A multi-model adaptive controller for a nonlinear system, the controller is composed of a nonlinear robust indirect adaptive controller, a nonlinear neural network adaptive controller and a switching mechanism, wherein one controller is The nonlinear robust indirect adaptive controller, the other is the nonlinear neural network indirect adaptive controller, the input of the controlled object is generated by selecting between the two controllers by the switching mechanism, the output of the controlled object and the two controllers There is a closed-loop negative feedback between them, and the relationship between the controlled object and the model output of the two indirect adaptive controllers is set as a subtraction relationship, and the model error is used to adjust the parameters of the model and the weights of the neural network.

The nonlinear robust indirect adaptive controller includes a nonlinear robust adaptive model and a nonlinear controller, the nonlinear robust adaptive model consists of two parts, linear and nonlinear, and the linear part corresponds to the system after linearization The linear part of , the nonlinear part is represented by the norm of the regression vector with coefficients, which is used to compensate the nonlinear part of the system after linearization. The regression vector consists of the input and output variables of the system at past moments. The adaptive parameter is the coefficient of the linear regression vector, and the adaptive law adopts the projection adaptive law. The nonlinear controller uses a one-step look-ahead controller.

The nonlinear neural network adaptive controller includes a nonlinear neural network adaptive model and a nonlinear neural network adaptive controller. The nonlinear neural network adaptive model consists of a linear part and a nonlinear neural network part. The coefficients of the linear part can be updated in any way as adaptive parameters. The nonlinear neural network adopts BP network and uses the error backpropagation method to train.

In the switching mechanism, a performance index is first designed, and the performance index includes a cumulative error part and a transient error part. At each control moment, the performance index of each controller is calculated, and the controller with a smaller performance index is selected to generate the control input at the next moment, which can realize smooth switching and improve the transient performance of the system.

The steps included in the control method of the multi-model adaptive controller of the above-mentioned nonlinear system are as follows:

S1: System initialization: Randomly initialize the parameters of the nonlinear robust adaptive model, randomly initialize the parameters of the nonlinear neural network model and the weights of the neural network, these parameters can be determined by certain prior knowledge;

S2: At k=0, the output of the object is 0; at k≠0, the output of the 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 system set value; the actual output and the nonlinear robustness The output of the adaptive model is subtracted to obtain the model error e ₁ , and the output of the nonlinear neural network model is subtracted to obtain the model error e ₂ ;

S3: the control error e _c is used as the input of the nonlinear robust adaptive controller and the nonlinear neural network adaptive controller, and the two controllers generate control quantities u ₁ and u ₂ respectively;

S4: Calculate the values of performance indicators C ₁ and C ₂ according to the model errors e ₁ and e ₂ , and select the input u _i generated by the controller with a smaller performance index value as the control input u of the controlled object and the two models;

S5: Using the model errors e1 and _e2 to update the parameters and weights of the nonlinear robust adaptive model and the nonlinear neural network adaptive model respectively _;

S6: go to step S2.

Compared with the prior art, the beneficial effects of the present invention are as follows:

From the multi-model adaptive control method of the present invention, it can be seen that a nonlinear robust adaptive model is included in the nonlinear robust adaptive controller, and an adaptive nonlinear adaptive model is added to the original linear adaptive model. The compensation term relaxes the restriction of the nonlinear term of the controlled object from linear bounded to zero-order close to bounded, which greatly expands the scope of application of the multi-model adaptive controller. A nonlinear system controlled solely by a nonlinear robust adaptive controller can be shown to be stable and convergent.

A nonlinear neural network model of the system is included in the nonlinear neural network adaptive controller. According to the universal approximation theorem of the neural network, this model can approach the real output of the system with arbitrary precision, which makes the control method of the present invention different from the traditional multi-model Compared with the adaptive control method, it has higher precision. The design of the controller adopts the idea of one-step advanced control, and the calculation amount is small, which can improve the calculation speed of the system.

Through the design of the switching mechanism, the controller of the system is switched between the nonlinear robust adaptive controller and the nonlinear neural network adaptive controller, and the controller with the smaller performance index value can be selected as the control input of the current system , which can reduce the transient error of the system. The performance index includes a cumulative term of error, which can prevent the system from frequently switching between two controllers and make the system output smoother. Due to the stability of the nonlinear robust adaptive control system, the present invention uses a nonlinear robust adaptive controller and a nonlinear neural network adaptive controller to switch, which can prove that the multi-model adaptive controller of the present invention has stability and convergence.

Description of drawings

Fig. 1 is the closed feedback control system block diagram of the present invention's design multi-model neural network adaptive controller;

Fig. 2 is a structural block diagram of a nonlinear robust indirect adaptive controller;

Fig. 3 is a structural block diagram of a nonlinear neural network indirect adaptive controller;

Fig. 4 is the structural block diagram of neural network;

Fig. 5 is the flowchart of switching mechanism;

Figure 6(1) and Figure 6(2) are the output curve and input curve of the controller of the present invention respectively.

specific implementation method

Below in conjunction with accompanying drawing and example, further illustrate the present invention.

As shown in Fig. 1, the controller designed by the multi-model adaptive control method of the present invention is composed of a nonlinear robust adaptive controller, a nonlinear neural network adaptive controller and a switching mechanism. In the figure, r(t+1) is the tracking reference signal of the system, u(t) is the input of the controlled object, and y(t+1) is the output of the controlled object. Nonlinear Robust Indirect Adaptive Controllers Containing Nonlinear Robust Adaptive Models and nonlinear robust adaptive controller is the controller Output, is a model Output. Nonlinear Neural Network Indirect Adaptive Controller Contains Nonlinear Neural Network Adaptive Model and nonlinear neural network adaptive controller is the controller Output, is a model Output. u(t) is determined by the switching mechanism in and Choose between.

The present invention is aimed at the nonlinear discrete-time system of the following structure

$\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; :</span>\left\{\begin{array}{c}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; G</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

where u(t), y(t)∈R are the input and output of the system respectively, x(t)∈R ⁿ is an n-dimensional state vector, F( ), G( ) are smooth nonlinear functions .

The system in a neighborhood of the origin can be represented by the following nonlinear model:

$\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

In the formula, a _i , i=0,…,n _a -1; b _j ,j=0,…,n _b are unknown parameters of the system, n _a , n _b are the order of the system, w(t)= [y(t),...,y(tn _a +1),u(t),...,u(tn _b )] ^T is a regression vector composed of system data.

The following assumptions are made for the above system (2):

A1. The order n _a and n _b of the system are known;

A2. Parameters a _i , i=0,...,n _a -1, b _j , j=0,...,n _b change in a compact set Ω;

A3. The system has a globally consistent and asymptotically stable zero dynamic system, so that the growth rate of any input sequence of the system does not exceed the growth rate of its corresponding output sequence;

A4. There is a known constant 0≤μ<∞, so that the function f(w(t)) is is zero-order nearly bounded, that is, satisfies in $\stackrel{\&OverBar;}{w}\left(t\right)={[\mathrm{the\; y}\left(t\right),...,\mathrm{the\; y}(t-{\mathrm{no}}_a+1),u(t-1),...,u(t-{\mathrm{no}}_b)]}^T,$ $g\left(\stackrel{\&OverBar;}{w}\left(t\right)\right)=\mathrm{\&lambda;}\left|\right|\stackrel{\&OverBar;}{w}\left(t\right)\left|\right|$ is a nonlinear function, and λ is an unknown arbitrary constant.

Figure 2 shows the structure diagram of nonlinear robust indirect adaptive control. The controller consists of two parts: nonlinear robust adaptive model and nonlinear adaptive controller. First, the nonlinear robust adaptive model of the plant is designed as

$\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>$

(3)

$<span\; class="notranslate">\; =</span>{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

In the formula $\underset{1}{\mathrm{\&theta;}}\left(t\right)={[\underset{1}{a_0}\left(t\right),...,\underset{1}{a_{{\mathrm{no}}_a-1}}\left(t\right),\underset{1}{b_0}\left(t\right),...,\underset{1}{b_{{\mathrm{no}}_b}}\left(t\right)]}^T$ and is a model The parameters at time t, let (2) formula can be obtained. At any system time t, by the model gives an estimate of the output of the system as The model error of the nonlinear robust adaptive model can be obtained by making a difference between the estimated value and the real value output by the system which is According to the model error The following robust adaptive identification algorithm with dead zone is used to update the model parameters:

$\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; \&psi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

In the formula, $\underset{1}{h}\left(t\right)=\left\{\begin{array}{cc}\frac{1}{2}& \left|\underset{1}{e}\left(t\right)\right|\&Greater\; Equal;2\mathrm{\&mu;}\\ 0& \mathrm{otherwise}\end{array}\right..$

At each system time t, according to the idea of one-step advance control, the parameters of the nonlinear robust adaptive model to design nonlinear adaptive controllers

$\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; b</span>}}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}}}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&psi;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

In the formula, $\underset{1}{\overset{^}{\stackrel{\&OverBar;}{\mathrm{\&delta;}}}}\left(t\right)={[{\underset{1}{\overset{^}{\stackrel{\&OverBar;}{\mathrm{\&theta;}}}}}^T\left(t\right),\underset{1}{\overset{^}{\mathrm{\&lambda;}}}\left(t\right)]}^T,$ $\underset{1}{\overset{^}{\stackrel{\&OverBar;}{\mathrm{\&theta;}}}}\left(t\right)={[\underset{1}{{\hat{a}}_0}\left(t\right),...,\underset{1}{{\hat{a}}_{{\mathrm{no}}_a-1}}\left(t\right),\underset{1}{{\hat{b}}_1}\left(t\right),...,\underset{1}{{\hat{b}}_{{\mathrm{no}}_b}}\left(t\right)]}^T,$ $\stackrel{\&OverBar;}{\mathrm{\&psi;}}\left(t\right)={[{\stackrel{\&OverBar;}{w}}^T\left(t\right),|\left|\stackrel{\&OverBar;}{w}\left(t\right)\right|\left|\right]}^T.$

Figure 3 shows the structure diagram of the nonlinear neural network indirect adaptive controller. The controller includes two parts: nonlinear neural network adaptive model and nonlinear neural network controller.

Design a Nonlinear Neural Network Identification Model for Nonlinear Controlled Objects

$\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

In the formula, $\underset{2}{\mathrm{\&theta;}}\left(t\right)={[\underset{2}{a_0}\left(t\right),...,\underset{2}{a_{{\mathrm{no}}_a-1}}\left(t\right),\underset{2}{b_0}\left(t\right),...,\underset{2}{b_{{\mathrm{no}}_b}}\left(t\right)]}^T$ is a model parameters, is a bounded nonlinear function approximation represented by a neural network, W(t) is the weight coefficient of the neural network, and the coefficient and W(t) in a pre-defined compact set S. is right at time t discernment, Update it in the following way:

$\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

In the formula, $\underset{2}{e}\left(t\right)=\mathrm{the\; y}\left(t\right)-\underset{2}{\overset{^}{\mathrm{the\; y}}}\left(t\right),$ $\underset{2}{h}\left(t\right)=\left\{\begin{array}{cc}\frac{1}{2}& \left|\underset{2}{e}\left(t\right)\right|\&Greater\; Equal;2\mathrm{\&mu;}\\ 0& \mathrm{otherwise}\end{array}\right..$ if $\underset{2}{{\hat{b}}_0}\left(t\right)<b_{\min },$ order $\underset{2}{{\hat{b}}_0}\left(t\right)=b_{\min }.$

According to the nonlinear neural network model The nonlinear control law of the system can be obtained as:

$\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; b</span>}}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{<span\; class="notranslate">\; ^</span>}{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

In the formula, $\underset{2}{\overset{^}{\stackrel{\&OverBar;}{\mathrm{\&theta;}}}}\left(t\right)={[\underset{2}{{\hat{a}}_0}\left(t\right),...,\underset{2}{{\hat{a}}_{{\mathrm{no}}_a-1}}\left(t\right),\underset{2}{{\hat{b}}_1}\left(t\right),...,\underset{2}{{\hat{b}}_{{\mathrm{no}}_b}}\left(t\right)]}^T.$

The structural diagram of the neural network shown in Figure 4 is a BP neural network with three layers of neurons, including an input layer, a hidden layer and an output layer. Each neuron between the upper and lower layers is fully connected, and there is no connection between neurons in the same layer. The connection weight l _ij from the input layer to the middle layer, i=1,2,...,n _a +n _b -2,j=1,2,...,p; the connection weight v _j1 from the hidden layer to the output layer, j= 1,2,...,p; the output threshold τ _j of each unit in the hidden layer, j=1,2,...,p; the output threshold of the output layer unit is γ ₁ ; the parameter k=1,2,...,m. Enter as $\stackrel{\&OverBar;}{w}\left(t\right)={[\mathrm{the\; y}\left(t\right),...,\mathrm{the\; y}(t-{\mathrm{no}}_a+1),u(t-1),...,u(t-{\mathrm{no}}_b)]}^T.$

The input s _j of each neuron in the hidden layer is: ${\mathrm{the\; s}}_j=\underset{i=1}{\overset{\mathrm{no}}{\mathrm{\&Sigma;}}}{l}_{\mathrm{ij}}{\stackrel{\&OverBar;}{w}}_i-{\mathrm{\&tau;}}_j,j=\mathrm{1,2},...,p.$ Use s _j to calculate the output b _j of each neuron in the hidden layer through the transfer function: b _j =g(s _j ),j=1,2,...,p. Using the output b _j of the hidden layer, the connection weight v _j1 and Threshold γ ₁ calculates the output L _t of neurons in the output layer as: The response of the neurons in the output layer is then calculated by the transfer function for: Using connection weight v _j1 , error and the output b _j of the hidden layer, calculate the error d _j (t) of each neuron in the hidden layer.

${\mathrm{<span\; class="notranslate">\; d</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; [</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ]</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>$

use output error Use the output b _j of each neuron in the hidden layer to modify the connection weight v _j1 and threshold γ ₁ :

${\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&kappa;</span>}\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$

${\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; \&gamma;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&kappa;</span>}\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>$

j=1,2,...,p,0<κ<1

Using the error d _j (t) of neurons in the hidden layer, the input of the input layer

$\stackrel{\&OverBar;}{w}\left(t\right)={[\mathrm{the\; y}\left(t\right),...,\mathrm{the\; y}(t-{\mathrm{no}}_a+1),u(t-1),...,u(t-{\mathrm{no}}_b)]}^T$ To modify the connection weight l _ij and the threshold τ _j :

τ _j =τ _j +σd _j (t).

i=1,2,...,n _a +n _b -2,j=1,2,...,p,0<σ<1

Figure 5 shows the flow chart of the switching mechanism. Firstly, performance indexes are designed for nonlinear robust indirect adaptive control and nonlinear neural network indirect adaptive control respectively. and Its calculation method is as follows:

$\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; t</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; [</span>{\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; 4</span>}\mathrm{<span\; class="notranslate">\; \&mu;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; c</span>}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; N</span>}}{\overset{\mathrm{<span\; class="notranslate">\; t</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; [</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>{\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

In the formula $\underset{\mathrm{the\; s}}{e}\left(t\right)=\mathrm{the\; y}\left(t\right)-\underset{\mathrm{the\; s}}{\overset{^}{\mathrm{the\; y}}}\left(t\right),$ $\underset{\mathrm{the\; s}}{h}\left(t\right)=\left\{\begin{array}{cc}\frac{1}{2}& \left|\underset{\mathrm{the\; s}}{e}\left(t\right)\right|>2\mathrm{\&mu;}\\ 0& \mathrm{otherwise}\end{array}\right.,$ μ≥0, N is a predefined integer, c≥0 is a constant. Judging the value of the performance index of the two controllers, and assigning the controller with the smaller performance index value to the control input u(t) of the system, namely:

$\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>& \underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\\ \underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; u</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>& {\mathrm{<span\; class="notranslate">\; J</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\underset{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span>\end{array}\right.$

According to the above discussion, the specific real-time online control steps of the multi-model adaptive control method of the present invention are as follows:

S1: System initialization: random initialization model and The parameters and neural network can be determined according to prior knowledge; (the neural network is set to a single hidden layer, and the number of neurons in the hidden layer is usually set to 6-10);

S2: At time t=0, the output of the system is zero, that is, y(0)=0; when t≠0, the real output value y(t) of the system is given by the controlled object of the system, and the model and gives the estimated output of the model respectively and The estimated error of the calculation model is and $\underset{2}{e}\left(t\right)=\mathrm{the\; y}\left(t\right)-\underset{2}{\mathrm{the\; y}}\left(t\right);$

S3: Calculate the control error e(t) of the system from the reference input r(t) of the system and the real output y(t) of the system;

S4: Utilize the model and parameters to design the controller and According to equations (5) (8), the output values of the nonlinear robust controller and the nonlinear neural network controller are respectively calculated from the control error e(t) of the system and

S5: Calculate the performance index of each controller from the model estimation error and The controller u _i (t) with a smaller value of the performance index is selected by the switching mechanism (11) as the control input u(t) of the controlled object;

S6: Estimated error by model and According to their respective adaptive laws (4) and (7), the models are updated separately and The parameters of and the weights of the neural network;

S7: Go back to step S2.

Figure 6 (1) and (2) are the input and output curves of the control system of the present invention respectively. It can be seen that the multi-model adaptive controller can track the reference sinusoid very well, and the control input is relatively gentle, which is easy to implement.

Claims (7)

1. the multi-model Adaptive Control device of a nonlinear system, it is characterised in that this controller is by two indirect self-adaptives Controller and a switching mechanism composition, wherein, a controller is non linear robust Indirect adaptive control device, and another is Nonlinear neural network Indirect adaptive control device, the input of controlled device is selected by switching mechanism to produce between the two controllers Raw, there are a close loop negative feedback, controlled device and two Indirect adaptive controls between the output of controlled device and two controllers Being set to subtract each other relation between the model output of device, model error is for adjusting the parameter of model and the weights of neutral net；

Non linear robust Indirect adaptive control device and nonlinear neural network Indirect adaptive control device separately design performance and refer to MarkWithIts computational methods are as follows:

In formulaμ >=0, N are predefined integers, and c >=0 is One constant, y (t) is the output signal of system,Being the estimate of system output signal, w (i-1) is system input and output The regression vector of signal composition；

The controller that the value of performance indications is less is selected to input as the control of controlled device by switching mechanism.

2. the multi-model Adaptive Control device of nonlinear system according to claim 1, it is characterised in that described non-linear Robust Indirect adaptive control device includes a non linear robust adaptive model and a gamma controller, non linear robust Adaptive model is by increasing a compensation term to mission nonlinear item on the basis of linear model, it is ensured that when system non-thread When the restrictive condition of property item is loosened to zeroth order close to bounded, the Identification Errors of this model also asymptotic can be less than a normal number.

3. the multi-model Adaptive Control device of nonlinear system according to claim 1, it is characterised in that described non-linear Neutral net Indirect adaptive control device includes a nonlinear neural network adaptive model and a nonlinear neural network Controller, nonlinear neural network adaptive model passes through on-line tuning neural network weight, it is thus achieved that the estimation to controlled device Output.

4. the multi-model Adaptive Control device of nonlinear system according to claim 2, it is characterised in that described nerve Network self-adapting model contains an input layer, a hidden layer and an output layer.

5. the multi-model Adaptive Control device of nonlinear system according to claim 4, it is characterised in that described nerve Containing 6-10 neuron in the hidden layer of network self-adapting model, output layer has a neuron.

6. the control method for described multi-model Adaptive Control device arbitrary in claim 1 to 5, its feature exists In the step of this control method is as follows:

S1: system initialization: the parameter of random initializtion non linear robust adaptive model, random initializtion non-linear neural net The parameter of network model and the weights of neutral net, these parameters can be determined by certain priori；

In the S2:k=0 moment, object is output as 0；In k ≠ 0 moment, object is output as the real output value of system, sets with system Definite value makees the control error e that difference obtains system_c；Actual output obtains model with the output work difference of non linear robust adaptive model Error e₁, make difference with nonlinear neural network model and obtain model error e₂；

S3: error e will be controlled_cDefeated as non linear robust adaptive controller and nonlinear neural network adaptive controller Enter, two controllers are produced controlled quentity controlled variable u respectively₁And u₂；

S4: according to model error e₁And e₂Carry out calculation of performance indicators C₁And C₂Value, select the less controller of performance index value to produce Raw input u_i, as the control input u of controlled device and two models,

S5: utilize model error e₁And e₂Update non linear robust adaptive model and nonlinear neural network adaptive mode respectively The parameter of type and weights；

S6: forward step S2 to.

7. the control method of the multi-model Adaptive Control device of nonlinear system according to claim 6, it is characterised in that In described step S4, in switching mechanism, first one performance indications of design, this performance indications comprise an accumulated error part With a transient error part, control the moment at each, calculate the performance indications of each controller, select performance indications relatively Little controller produces the control input of subsequent time.