CN106773712A - Double feedback robust self-adaptation control methods and its Control system architecture - Google Patents

Abstract

The invention discloses a kind of pair of feedback robust self-adaptation control method and its Control system architecture, L is inherited₁The ability of the powerful treatment uncertain factor of self-adaptation control method, but by the built-in controll plant dynamic dead time delay model of controller, using the forecast function of model, overcoming L₁Adaptive controller time lag allowance is small, be not applied for the shortcoming of big dead time delay process；The present invention utilizes L simultaneously₁Self Adaptive Control processes the ability of uncertain factor, using a kind of adaptive law of new Dual-loop feedback control strategy, reduces the requirement to built-in dynamic dead time delay model accuracy.Compared with other System design based on model methods, the present invention can have the built-in dynamic dead time delay model of larger modeling error using dynamic characteristic and time lag characteristic, and superiority is had more in terms of robustness and stability.Thus the present invention can be directly applied for big dead time delay uncertainty thermal technology and Chemical Engineering Process Control, including without obvious dead time delay and probabilistic controlled process.

Description

Double-feedback robust self-adaptive control method and control system structure thereof

Technical Field

The invention belongs to the field of thermal power engineering and automatic control, and particularly relates to a double-feedback robust self-adaptive control method and a control system structure thereof, which are suitable for a large dead time lag uncertainty process.

Background

At present, the operation of a thermal system represented by a thermal power generating unit faces more and more challenges, such as rapid load tracking, large-range variable working condition operation, unknown uncertainty caused by fuel calorific value change, and immeasurable disturbance caused by smoke emission control. To deal with these problems, the thermal power unit control should have fast, adaptive and robust stability performance. L is₁The self-adaptive control is a novel rapid robust self-adaptive control method, and compared with other advanced control such as predictive control and the like, the self-adaptive control has strong capability in the aspect of processing uncertainty problems such as nonlinearity, undetectable disturbance, model mismatching and the like, so that the self-adaptive control has an application prospect in a thermal power generating unit bearing power grid peak regulation and frequency modulation tasks. However, L₁The self-adaptive controller has smaller time-lag margin and is easy to disperse when controlling a large pure time-lag object, and a large amount of pure time-lag phenomena exist in the thermal engineering and chemical engineering processes, so that L₁Adaptive controllers are currently not suitable for use in a large number of such processes.

Disclosure of Invention

The technical problem is as follows: the invention provides a double-feedback robust self-adaptive control method capable of implementing rapid, robust, self-adaptive and stable control on a large dead time uncertainty process, aiming at solving the problem that the existing control method is difficult to simultaneously process three aspects of large dead time, nonlinearity and unknown disturbance.

The technical scheme is as follows: the invention discloses a double-feedback robust self-adaptive control method, which comprises the following steps:

step 1: initializing the structure, parameters and variables of the double-feedback robust adaptive controller, wherein the initialization comprises the initialization of an expected reference system, a state estimator and a built-in dynamic dead time lag model, and the initialization of the built-in dynamic dead time lag model comprises the initialization of a dead time-free dynamic model and a dead time lag model;

step 2: at each sampling moment, the double-feedback robust adaptive controller outputs a bounded reference input vector r (t) serving as a set value and a state vector without dead time lag output from a dynamic model without dead time lagDead-time-lag-bearing state vector output from dead-time-lag modelState vector estimator output from state estimatorAnd sampling a measurable state vector x (t) output from the controlled process, calculating a state prediction bias by a dual feedback loop

And step 3: predicting deviation of stateInputting adaptive law, calculating uncertainty estimator of controlled process

And 4, step 4: estimating an uncertaintySending the control signals into a low-pass filter and a control law module which are connected in series, and calculating the control quantity u (t) at the current moment;

and 5: sending the control quantity u (t) to a controlled object executing mechanism, and driving the controlled object to track a set value along the input and output tracks of the expected reference system; meanwhile, the control quantity u (t) is sent to a built-in dynamic dead time model, and the built-in dynamic dead time model is calculated in the next sampling period; meanwhile, the control quantity u (t) is also sent to a state estimator, and state estimation is carried out in the next sampling period; and returning to the step 2, and controlling the next sampling period.

Further, to implement step 1, the desired reference system and state estimator are described and designed for the controlled process, and the steps are as follows:

step 1-1: describing a large dead time lag uncertainty controlled process as a nonlinear state space equation with output dead time lag; the equation consists of an unmeasured state equation, an internal unmodeled state equation, a measurable state equation with pure time lag and an output equation:

wherein x is_D(t)＝[x_D1(t)…x_Dn(t)]^T∈RⁿAs an undetectable state vector, x₀Is x_D(t) initial vector value, x_Di(t) is x_D(t), i ═ 1, …, n;

x(t)＝[x₁(t)…x_n(t)]^T∈Rⁿis a measurable state vector, the component x of which_i(T) has a pure time lag T_di≥0，i＝1,…,n；

z (t) represents an unmodeled internal unmeasured state vector, the unknown function g (z, x)_DT) represents the dynamic behavior of z (t), z₀An initial vector value of z (t);

control quantity u (t) ∈ R^mInput vector for controlled process, y (t) ∈ R^mOutputting a vector for the controlled process; m and n represent the dimensions of the vector;

f(x_Dz, t) is an unknown function expressing the nonlinear dynamic characteristics of the object; coefficient matrix A_D,B_D,C_DRepresenting the desired dynamic characteristics of a stable closed-loop system, D is an output coefficient matrix；

Step 1-2: desired dynamic characteristics A described according to step 1-1_D,B_D,C_DDefining a desired reference system with dead time lag, described by the following state space equations:

wherein,in order for the reference state vector to be undetectable,is composed of1, …, n;

x_des(t)＝[x_des,1(t),…,x_des,n(t)]∈Rⁿis a pure time-lagged reference state vector, the component x of which_des,i(T) has a pure time lag T_di≥0，i＝1,…,n；

y_des(t)∈R^mIs a reference output vector; coefficient matrixr(t)∈R^mInputting a vector for a given bounded reference;

step 1-3: desired dynamic characteristics A described according to step 1-1_D,B_D,C_DDefining the state estimator as:

wherein the state vector estimatorAnd output vector estimatorEstimates of a measurable state vector x (t) and an output vector y (t) of the controlled object, respectively,is an uncertainty estimator of the system, x₀Is composed ofThe initial vector value of (a), the control quantity u (t) ∈ R^mAlso the input vector to the state estimator.

Further, step 1 also includes steps 1-4: in order to realize the double-feedback strategy of the invention, a built-in dynamic dead time lag model of the controlled object is required to be added in the double-feedback robust self-adaptive controller, and the built-in dynamic dead time lag model consists of a dead time lag-free dynamic model and a dead time lag model.

Furthermore, the built-in dynamic dead time lag model can be described as a dead time lag-free dynamic model and a dead time lag model; a dynamic model without dead time lag can be described in the form of the following equation of state:

wherein,for a state vector without a pure time lag,is composed of1, …, n; a. the_m、B_mIs a matrix of coefficients, x₀Is composed ofThe initial vector value of (a), the control quantity u (t) ∈ R^mAn input vector of a built-in dynamic dead time lag model;

the pure time lag model can be described in the form of the following pure time lag equation and output equation:

wherein the input quantity of the dead time lag model is a state vector component without dead time lagi＝1,…,n；Is a state vector with dead time lag, its componentsWith dead time T_mi≥0,i＝1,…,n；Outputting vectors for the built-in dynamic dead time lag model; c_m、D_mIs a matrix of coefficients.

Further, to implement step 2, a state prediction bias is calculated based on the dual feedback state vectorThe method comprises the following steps:

step 2-1: calculating the inner loop feedback deviation

Wherein,is the state vector of the state estimator,the state vector is a state vector without dead time lag of a built-in dynamic dead time lag model;

step 2-2: calculating an outer loop feedback offset

Wherein,the state vector with dead time lag is a built-in dynamic dead time lag model, and x (t) is a measurable state vector of a controlled process;

step 2-3: calculating a state prediction biasThe feedback error is obtained by adding the inner loop feedback error and the outer loop feedback error.

Further, to implement step 3, the states are predicted to be biasedInputting the adaptive law, calculating the uncertainty estimatorAdaptive law for predicting deviations from stateAs input, with an uncertainty estimatorIs output as followsPiecewise linear function:

wherein,

t >0 denotes the sampling period and i denotes the sampling sequence number.

Further, in step 4, the control amount u (t) is generated by the following control law:

wherein the coefficient matrix

F(s) is a low pass filter with a steady state gain of 1, s is the Laplace transform operator, r(s) anda bounded reference input vector r (t) and an uncertainty estimator, respectivelyOf laplace transform, L^-1[·]Representing an inverse laplace transform.

The invention also discloses a control system structure of the double-feedback robust self-adaptive control method, which comprises a control law, a dynamic model without dead time lag, a dead time lag model, a state estimator, a first adder for forming an inner loop feedback deviation signal, a second adder for forming an outer loop feedback deviation signal, a third adder for forming a state prediction deviation signal, a self-adaptive law module and a low-pass filter; the input end of the control law inputs an input signal, the output signal of the control law is simultaneously sent to a controlled object, a dead time-lag-free dynamic model and a state estimator, the controlled object is driven by the output signal of the control law, the output end of the controlled object outputs a measurable state signal, one path of the output signal of the dead time-lag-free dynamic model is sent to the input end of the dead time-lag model, the other path of the output signal of the dead time-lag-free dynamic model is sent to a first adder, the output signal of the dead time model and the measurable state signal are sent to a second adder to obtain and output an outer loop feedback deviation signal, the control quantity signal output by the control law and the output signal of the adaptive law module are used as input signals to enter the state estimator, the output signal of the state estimator is sent to the first adder to output an inner loop feedback deviation signal, the inner loop feedback deviation signal output by the first adder and the outer loop feedback deviation signal output by the second The state prediction deviation signal is sent to an adaptive law module, one path of the output end of the adaptive law module is sent to a state estimator, the other path of the output end of the adaptive law module enters a control law through a low-pass filter connected with the adaptive law module in series, and the input signal of the control law also comprises a bounded reference input signal.

Has the advantages that: the invention inherits L₁The self-adaptive control method has strong capability of processing uncertain factors, but overcomes the defect of L by arranging a controlled object dynamic dead time delay model in the controller and utilizing the prediction function of the model₁The self-adaptive controller has the defects of small time lag allowance and incapability of being suitable for a large pure time lag process; while the present invention utilizes L₁The capability of self-adaptive control for processing uncertain factors adopts a self-adaptive law of a new double closed-loop feedback strategy, and reduces the requirement on the precision of a built-in dynamic dead time lag model. Compared with other control methods based on models, the method can use the built-in dynamic dead time lag model with larger modeling errors in both dynamic characteristics and time lag characteristics, and is more robust and stableHas the advantages. Therefore, the invention can be directly applied to the control of large dead time uncertainty thermotechnical and chemical processes, including controlled processes without obvious dead time and uncertainty.

Drawings

FIG. 1 is a flow chart of a control process of the present invention;

FIG. 2 is a block diagram of the control system of the present invention;

FIG. 3 is L₁Adaptive control system architecture;

FIG. 4 shows L in an embodiment₁Step response curve of adaptive control system:

wherein FIG. 4-1 shows a step response curve of steam pressure; FIG. 4-2 shows a step response curve for power; 4-3 show step response curves for drum water levels; 4-4 show step response curves for fuel flow, steam valve opening, and feedwater flow;

fig. 5 is a step response curve of the control system of the present invention in an embodiment (dead time accurate with a built-in dynamic dead time model):

wherein FIG. 5-1 shows a step response curve for steam pressure; FIG. 5-2 shows a step response curve for power; 5-3 show step response curves for drum water levels; 5-4 show step response curves for fuel flow, steam valve opening, and feedwater flow;

FIG. 6 is a step response curve (dead time offset for built-in dynamic dead time model) for a control system of the present invention in an exemplary embodiment:

wherein FIG. 6-1 shows a step response curve for steam pressure; FIG. 6-2 shows a step response curve for power; 6-3 show a step response curve for drum level; fig. 6-4 show step response curves for fuel flow, steam valve opening, and feedwater flow.

Detailed Description

And the model which is widely used as a test object of the thermal control algorithm internationally, namely R.D. Bell and K.J.The invention is explained by taking the built 160MW unit machine furnace model as an example and referring to the attached drawings.

The model has no obvious pure time lag, and a pure time lag link is added on the model for implementing and verifying the invention. Adding dead time does not affect the dynamics of the model, i.e., for a given input change, the model only delays its output change by a dead time. The 160MW unit furnace model with increased dead time is as follows:

q_e(t)＝[0.854u₂(t)-0.147]x_D1(t)+45.59u₁(t)-2.514u₃(t)-2.096， (2)

wherein the controlled quantity is the fuel flow u₁Value of [0,1]Opening u of steam valve₂Value of [0,1]And the feed water flow u₃Value of [0,1]All are normalized variables and are constrained by the following conditions:

0≤u_i≤1,i＝1,2,3；

wherein, the state variable drum pressure x_D1(unit is kg/cm)²) Generated power x_D2(in MW) and drum liquid density x_D3(unit is kg/cm)³) Being unmeasurable, pure time-lagging state variable drum pressure x₁Generated power x₂And drum level deviation x₃(in m) is measurable, one-to-one corresponding to the output variable y₁、y₂、y₃Respectively having dead time-lag time T_d1、T_d2、T_d3(unit s).

The machine furnace model has obvious pure time lag characteristic and has strong nonlinearity and uncertainty when operating in a large-range variable working condition.

A control system is designed for the model by applying the double-feedback robust self-adaptive control method provided by the invention.

Step 1: initializing the structure, parameters and variables of the double-feedback robust adaptive controller, wherein the initialization comprises the initialization of an expected reference system, a state estimator and a built-in dynamic dead time lag model, and the initialization of the built-in dynamic dead time lag model comprises the initialization of a dead time-free dynamic model and a dead time lag model;

step 1-1: the 160MW unit furnace model to increase dead time is described as:

wherein x is_D(t)＝[x_D1(t)…x_Dn(t)]^T∈RⁿAs an undetectable state vector, x₀Is x_D(t) initial vector value, x_Di(t) is x_D(t), i ═ 1, …, n;

x(t)＝[x₁(t)…x_n(t)]^T∈Rⁿis a measurable state vector, the component x of which_i(T) has a pure time lag T_di≥0，i＝1,…,n；

z (t) represents an unmodeled internal unmeasured state vector, the unknown function g (z, x)_DT) represents the dynamic behavior of z (t), z₀An initial vector value of z (t);

control quantity u (t) ∈ R^mInput vector for controlled process, y (t) ∈ R^mOutputting a vector for the controlled process; m and n represent the dimensions of the vector, m-n-3 in this example;

f(x_Dz, T) is an unknown function for expressing the nonlinear dynamic characteristics of the object, D is an output coefficient matrix, the symbol ' ∈ ' represents ' belonging ', the ' (-) on the variable represents taking derivatives, and the superscript ' T ' on the variable represents the transpose of a matrix or a vector.

Matrix A_D,B_D,C_DThe expected dynamic characteristics of a stable closed loop system are described, and the values are as follows:

step 1-2: desired dynamic characteristics A described according to step 1-1_D,B_D,C_DDefining a desired reference system with dead time lag, described by the following state space equations:

wherein,in order for the reference state vector to be undetectable,is composed of1, …, n;

x_des(t)＝[x_des,1(t),…,x_des,n(t)]∈Rⁿis a pure time-lagged reference state vector, the component x of which_des,i(T) has a pure time lag T_di≥0,i＝1,…,n；

y_des(t)∈R^mIs a reference output vector; coefficient matrixWherein the superscript "-1" denotes the matrix inversion, R (t) ∈ R^mInputting a vector for a given bounded reference;

the double-feedback robust adaptive controller has the function of ensuring that the output vector y (t) of the controlled process tracks the reference output vector y of the expected reference system in a bounded way for any bounded reference input vector r (t)_des(t)；

Step 1-3: period described according to step 1-1Inspection of dynamic characteristics A_D,B_D,C_DDefining the state estimator as:

wherein the state vector estimatorAnd output vector estimatorEstimates of a measurable state vector x (t) and an output vector y (t) of the controlled object, respectively,is an uncertainty estimator of the system, x₀Is composed ofThe initial vector value of (a), the control quantity u (t) ∈ R^mAlso the input vector of the state estimator;

step 1-4: the method comprises the following steps of (1) embedding a dynamic dead time lag model of a controlled object in a controller: the built-in dynamic dead time lag model can be described as a dead time lag-free dynamic model and a dead time lag model. The present embodiment adopts a dead-time-lag-free dynamic model in the form of a state equation, which is described as follows:

wherein,without dead time lagThe state vector is then used to determine the state of the device,is composed of1, …, n; a. the_m、B_mIs a coefficient matrix; x is the number of₀Is composed ofThe initial vector value of (a), the control quantity u (t) ∈ R^mIs an input vector of a built-in dynamic dead time model.

The pure time lag model can be described in the form of the following pure time lag equation and output equation:

wherein the input quantity of the dead time lag model is a state vector component without dead time lagi＝1,…,n；Is a state vector with dead time lag, its componentsWith dead time T_mi≥0，i＝1,…,n；Outputting vectors for the built-in dynamic dead time lag model; c_m、D_mIs a matrix of coefficients.

Besides the built-in dynamic dead time lag model described by the state equation, the dead time lag equation and the output equation, the built-in dynamic dead time lag model in other description forms is also applicable, and the performance of the controller is better the closer the used model and the controlled object are to the characteristic, so the built-in dynamic dead time lag models in various description forms belong to the protection scope of the invention.

Carrying out Taylor series expansion on the 160MW unit machine-made furnace model with the increased dead time lag at the working condition point x (0) ([ 108,66.65,0 ]), and obtaining the coefficient of the dynamic dead time lag model as follows:

step 2: calculating state prediction bias from dual feedback state vectorsThe method comprises the following steps:

step 2-1: calculating the inner loop feedback deviation

Wherein,is the state vector of the state estimator,the state vector is a state vector without dead time lag of a built-in dynamic dead time lag model;

step 2-2: calculating an outer loop feedback offset

Wherein,the state vector with dead time lag is a built-in dynamic dead time lag model, and x (t) is a measurable state vector of a controlled process;

step 2-3: calculating a state prediction biasThe feedback error is obtained by adding the inner loop feedback error and the outer loop feedback error.

And step 3: predicting deviation of stateInputting the adaptive law, calculating the uncertainty estimatorAdaptive law for predicting deviations from stateAs input, with an uncertainty estimatorThe following piecewise linear function for the output:

wherein,

i denotes a sampling sequence number, T >0 denotes a sampling period, and T is 0.01 s.

And 4, step 4: estimating an uncertaintyAnd (4) sending the control signals to a low-pass filter and a control law module which are connected in series, and calculating the control quantity u (t) at the current moment. The control quantity u (t) is generated by the following control law:

u(t)＝L^-1[u(s)], (11)

wherein

F(s) is a low pass filter with a steady state gain of 1, s is the Laplace transform operator, r(s) anda bounded reference input vector r (t) and an uncertainty estimator, respectivelyOf laplace transform, L^-1[·]Representing an inverse laplace transform.

And 5: sending the control quantity u (t) to a controlled object executing mechanism, and driving the controlled object to track a set value along the input and output tracks of the expected reference system; meanwhile, the control quantity u (t) is sent to a built-in dynamic dead time model, and the built-in dynamic dead time model is calculated in the next sampling period; meanwhile, the control quantity u (t) is sent to a state estimator, and state estimation is carried out in the next sampling period; and returning to the step 2, and controlling the next sampling period.

In this embodiment, to verify the effect of the invention, a large-scale load step simulation experiment is performed, so that the state vector setting value is stepped from the initial steady-state point x (0) ([ 108,66.65,0], to another steady-state point x (t) ([ 129.6,105.8,0 ]). For comparative effect, three different controller settings were used:

(1)L₁the structure of the adaptive controller is shown in figure 3. Make the dead time lag time of the machine furnace system as T_d1＝T_d2＝T_d3Fig. 4-1 shows a step response curve for steam pressure, fig. 4-2 shows a step response curve for power, fig. 4-3 shows a step response curve for drum level, and fig. 4-4 shows a step response curve for fuel flow, steam valve opening, and feedwater flow.

(2) The structure of the controller of the invention is shown in figure 2. Make the dead time lag time of the machine furnace system as T_d1＝60s，T_d2＝50s，T_d3Setting the dead time of the built-in dynamic dead time model as T as 40s_m1＝60s，T_m2＝50s，T_m3Fig. 5-1 shows a step response curve for steam pressure, fig. 5-2 shows a step response curve for power, fig. 5-3 shows a step response curve for drum level, and fig. 5-4 shows a step response curve for fuel flow, steam valve opening, and feedwater flow.

(3) The structure of the controller of the invention is shown in figure 2. Make the dead time lag time of the machine furnace system as T_d1＝60s，T_d2＝50s，T_d3Setting the dead time of the built-in dynamic dead time model as T as 40s_m1＝66s，T_m2＝55s，T_m344s, with a step response as shown in fig. 6, fig. 6-1 showing a step response curve for steam pressure, fig. 6-2 showing a step response curve for power, fig. 6-3 showing a step response curve for drum level, and fig. 6-4 showing a step response curve for fuel flow, steam valve opening, and feedwater flow.

(1) The step responses of (1) and (2) are shown in FIGS. 4 and 5, respectively, and demonstrate that the controller ratio L of the present invention is the same under the same conditions₁The range of time lags that the adaptive controller can control is much larger. (3) The step response is shown in figure 6, which shows that the controller of the invention can tolerate the larger difference between the time lag parameter of the built-in dynamic dead time lag model and the actual controlled process, and the controller of the invention can be used for the built-in dynamic dead time lag modelHas stronger robustness, thus being widely applied to actual large dead time lag uncertainty controlled processes, including thermal and chemical process control.

The above-described arrangements are only preferred embodiments of the present invention and are not exhaustive of the possible embodiments of the present invention. Any obvious modifications to the above would be included within the scope of the present invention, which is set forth in the following claims, and the scope of the invention is to be determined by a person of ordinary skill in the art.

Claims (8)

1. A double-feedback robust self-adaptive control method is characterized in that: the method comprises the following steps:

step 1: initializing the structure, parameters and variables of the double-feedback robust adaptive controller, wherein the initialization comprises the initialization of an expected reference system, a state estimator and a built-in dynamic dead time lag model, and the initialization of the built-in dynamic dead time lag model comprises the initialization of a dead time-free dynamic model and a dead time lag model;

step 2: at each sampling instant, the dual feedback robust adaptive controller pairs a bounded parameter as a setpointState vector without dead time lag output from dynamic model without dead time lag by considering input vector r (t)Dead-time-lag-bearing state vector output from dead-time-lag modelState vector estimator output from state estimatorAnd sampling a measurable state vector x (t) output from the controlled process, calculating a state prediction bias by a dual feedback loop

And step 3: predicting deviation of stateInputting adaptive law, calculating uncertainty estimator of controlled process

And 4, step 4: estimating an uncertaintySending the control signals into a low-pass filter and a control law module which are connected in series, and calculating the control quantity u (t) at the current moment;

and 5: sending the control quantity u (t) to a controlled object executing mechanism, and driving the controlled object to track a set value along the input and output tracks of the expected reference system; meanwhile, the control quantity u (t) is sent to a built-in dynamic dead time model, and the built-in dynamic dead time model is calculated in the next sampling period; meanwhile, the control quantity u (t) is also sent to a state estimator, and state estimation is carried out in the next sampling period; and returning to the step 2, and controlling the next sampling period.

2. The dual feedback robust adaptive control method according to claim 1, wherein: to implement step 1, the controlled process is described, and the desired reference system and state estimator are designed, as follows:

step 1-1: describing a large dead time lag uncertainty controlled process as a nonlinear state space equation with output dead time lag; the equation consists of an unmeasured state equation, an internal unmodeled state equation, a measurable state equation with pure time lag and an output equation:

$\begin{array}{c}{\stackrel{\·}{x}}_D\left(t\right)=A_D{x}_D\left(t\right)+B_{D}u\left(t\right)+f(x_D,z,t),x_D\left(0\right)=x_0,\\ \stackrel{\·}{z}\left(t\right)=g(z,x_D,t),z\left(0\right)=z_0,\\ x_i\left(t\right)=x_{Di}(t-T_{di}),i=1,\mathrm{...},n,\\ y\left(t\right)=C_D^{T}x\left(t\right)+Du\left(t\right),\end{array}---\left(4\right)$

wherein x is_D(t)＝[x_D1(t)…x_Dn(t)]^T∈RⁿAs an undetectable state vector, x₀Is x_D(t) initial vector value, x_Di(t) is x_D(t), i ═ 1, …, n;

x(t)＝[x₁(t)…x_n(t)]^T∈Rⁿis a measurable state vector, the component x of which_i(T) has a pure time lag T_di≥0，i＝1,…,n；

z (t) represents an unmodeled internal unmeasured state vector, the unknown function g (z, x)_DT) represents the dynamic behavior of z (t), z₀An initial vector value of z (t);

control quantity u (t) ∈ R^mInput vector for controlled process, y (t) ∈ R^mOutputting a vector for the controlled process, m and n representing the dimensions of the vector;

f(x_Dz, t) is an unknown function expressing the nonlinear dynamic characteristics of the object; coefficient matrix A_D,B_D,C_DRepresenting the desired dynamic characteristics of a stable closed loop system, D being the outputA coefficient matrix is obtained;

step 1-2: desired dynamic characteristics A described according to step 1-1_D,B_D,C_DDefining a desired reference system with dead time lag, described by the following state space equations:

$\begin{array}{c}{\stackrel{\·}{\stackrel{\‾}{x}}}_{des}\left(t\right)=A_D{\stackrel{\‾}{x}}_{des}\left(t\right)+B_D{K}_{g}r\left(t\right),\\ x_{des,i}\left(t\right)={\stackrel{\‾}{x}}_{des,i}(t-T_{di}),\\ y_{des}\left(t\right)=C_D^T{x}_{des}\left(t\right),\end{array}---\left(5\right)$

wherein,in order for the reference state vector to be undetectable,is composed of1, …, n;

x_des(t)＝[x_des,1(t),…,x_des,n(t)]∈Rⁿis a pure time-lagged reference state vector, the component x of which_des,i(T) has a pure time lag T_di≥0,i＝1,…,n；

y_des(t)∈R^mIs a reference output vector; coefficient matrixr(t)∈R^mInputting a vector for a given bounded reference;

step 1-3: desired dynamic characteristics A described according to step 1-1_D,B_D,C_DDefining the state estimator as:

$\stackrel{\·}{\hat{x}}\left(t\right)=A_D\hat{x}\left(t\right)+B_{D}u\left(t\right)+\widehat{\mathrm{\σ}}\left(t\right),\hat{x}\left(0\right)=x_0,$

$\hat{y}\left(t\right)=C_D^T\hat{x}\left(t\right)---\left(6\right)$

wherein the state vector estimatorAnd output vector estimatorEstimates of a measurable state vector x (t) and an output vector y (t) of the controlled object, respectively,is an uncertainty estimator of the system, x₀Also is thatThe initial vector value of (a), the control quantity u (t) ∈ R^mAlso the input vector to the state estimator.

3. A dual feedback robust adaptive control method according to claim 1 or 2, characterized in that: the step 1 also comprises steps 1-4: in order to realize the double-feedback strategy of the invention, a built-in dynamic dead time lag model of a controlled object is added in the double-feedback robust self-adaptive controller, and the built-in dynamic dead time lag model consists of a dead time lag-free dynamic model and a dead time lag model.

4. A dual feedback robust adaptive control method according to claim 3, characterized in that: the built-in dynamic dead time lag model can be described as a dead time lag-free dynamic model and a dead time lag model; a dynamic model without dead time lag can be described in the form of the following equation of state:

$\stackrel{\·}{\stackrel{\‾}{x}}\left(t\right)=A_m\stackrel{\‾}{x}\left(t\right)+B_{m}u\left(t\right),\stackrel{\‾}{x}\left(0\right)=x_0,---\left(7\right)$

wherein,for a state vector without a pure time lag,is composed of1, …, n; a. the_m、B_mIs a matrix of coefficients, x₀Is composed ofThe initial vector value of (a), the control quantity u (t) ∈ R^mAn input vector of a built-in dynamic dead time lag model;

the pure time lag model can be described in the form of the following pure time lag equation and output equation:

$\begin{array}{c}{\stackrel{\‾}{x}}_{ei}\left(t\right)={\stackrel{\‾}{x}}_i(t-T_{mi}),i=1,\mathrm{...},n,\\ \stackrel{\‾}{y}\left(t\right)=C_m^T{\stackrel{\‾}{x}}_e\left(t\right)+D_{m}u\left(t\right),\end{array}---\left(8\right)$

wherein the input quantity of the dead time lag model is a state vector component without dead time lagi＝1,…,n；

Is a state vector with dead time lag, its componentsWith dead time T_mi≥0,i＝1,…,n；Output vector, C, for a built-in dynamic dead-time model_m、D_mIs a matrix of coefficients.

5. A dual feedback robust adaptive control method according to claim 1, 2 or 4, wherein: to implement step 2, a state prediction bias is calculated based on the dual feedback state vectorThe method comprises the following steps:

step 2-1: calculating the inner loop feedback deviation

Wherein,is the state vector of the state estimator,the state vector is a state vector without dead time lag of a built-in dynamic dead time lag model;

step 2-2: calculating an outer loop feedback offset

Wherein,the state vector with dead time lag is a built-in dynamic dead time lag model, and x (t) is a measurable state vector of a controlled process;

step 2-3: calculating a state prediction biasThe feedback error is obtained by adding the inner loop feedback error and the outer loop feedback error.

6. The dual feedback robust adaptive control method according to claim 5, wherein: to implement step 3, the states are predicted to be biasedInputting the adaptive law, calculating the uncertainty estimatorAdaptive law for predicting deviations from stateAs input, with an uncertainty estimatorThe following piecewise linear function for the output:

$\widehat{\mathrm{\σ}}\left(t\right)=\widehat{\mathrm{\σ}}\left(iT\right),t\∈\[iT,(i+1)T),$

$\widehat{\mathrm{\σ}}\left(iT\right)=-{\mathrm{\Φ}}^{-1}\left(T\right)\mathrm{\μ}\left(iT\right),i=0,1,2,...,---\left(9\right)$

wherein,

t >0 denotes the sampling period and i denotes the sampling sequence number.

7. The dual feedback robust adaptive control method according to claim 6, wherein: in the step 4, the control quantity u (t) is generated by the following control law:

$u\left(s\right)=K_{g}r\left(s\right)-F\left(s\right)\widehat{\mathrm{\σ}}\left(s\right),$

u(t)＝L^-1[u(s)], (11)

wherein the coefficient matrix

F(s) is a low pass filter with a steady state gain of 1, s is the Laplace transform operator, r(s) anda bounded reference input vector r (t) and an uncertainty estimator, respectivelyOf laplace transform, L^-1[·]Representing an inverse laplace transform.

8. The control system architecture for a dual feedback robust adaptive control method as claimed in claim 1, wherein: the system comprises a control law, a dead time lag-free dynamic model, a dead time lag model, a state estimator, a first adder for forming an inner loop feedback deviation signal, a second adder for forming an outer loop feedback deviation signal, a third adder for forming a state prediction deviation signal, an adaptive law module and a low-pass filter; the input end of the control law inputs an input signal, the output signal of the control law is simultaneously sent to a controlled object, a dead time-lag-free dynamic model and a state estimator, the controlled object is driven by the output signal of the control law, the output end of the controlled object outputs a measurable state signal, one path of the output signal of the dead time-lag-free dynamic model is sent to the input end of the dead time-lag model, the other path of the output signal of the dead time-lag-free dynamic model is sent to a first adder, the output signal of the dead time model and the measurable state signal are sent to a second adder to obtain and output an outer loop feedback deviation signal, the control quantity signal output by the control law and the output signal of the adaptive law module are used as input signals to enter the state estimator, the output signal of the state estimator is sent to the first adder to output an inner loop feedback deviation signal, the inner loop feedback deviation signal output by the first adder and the outer loop feedback deviation signal output by the second The state prediction deviation signal is sent to an adaptive law module, one path of the output end of the adaptive law module is sent to a state estimator, the other path of the output end of the adaptive law module enters a control law through a low-pass filter connected with the adaptive law module in series, and the input signal of the control law also comprises a bounded reference input signal.