Summary of the invention
In order to solve prior art problem, provide one kind can be used in extending to C-A time delay transition probability when S-C all the present invention
The fault detection method of network control system under conditions of part is unknown.
The technical scheme is that a kind of network control system fault detection method, comprising the following steps:
Step 1:
It utilizesAnd μkSensor in network control system is respectively indicated to the time delay and controller of controller to actuator
Time delay, and respectively in finite aggregateWith N={ μmin, μmaxIn value, transfer is general
Rate matrix is respectively as follows: ∧=[λij], Π=[πrs], λijAnd πrsIt is defined as follows:
In formula:
ForIt enablesWherein,IfThenWithIt is further represented as
Wherein 1≤d≤nM, nMIt is the number for gathering M element;ForIt enables Wherein,WhenThenWithIt is further represented asWherein 1≤g≤nN, nNIt is set N element
Number;
Step 2:
To the network control system with time delay:
In formula: xk∈RnIt is system mode vector, yk∈RmIt is system output vector, dk∈RpIt is L2The external disturbance of bounded
Signal, fk∈RqIt is L2The fault-signal of bounded, Ap, Bp, Bd, Bf, CpIt is the permanent matrix of appropriate dimension, K is control to be designed
Device gain matrix;Fault Detection Filter is constructed in controller end:
In formula:It is filter status vector,It is filter output vector;rk∈RqFor residual error, L be to
Filter gain matrix is designed, V is residual error gain matrix;
Step 3:
Definition status evaluated error and residual error error signalrek=rk-fk;Define augmentation vector Obtain closed-loop system equation:
In above formula:
I1=[0 I]T∈R2n×n, I2=[I 0]T∈Rn×2n, I3=[0 I]T∈R(p×q);
For unknown parameter K and L in processing system, augmentation vector is defined:
Then closed-loop system is written as:
In formula:
WhereinThePiecemeal is unit battle array, remaining is zero gust,ThePiecemeal is unit battle array,
Remaining is zero gust;
Step 4:
K and L unknown in step 3 in order to obtain are solved:
Above-mentioned formula meets:
Provide controller gain matrix K and filter gain matrix L and minimum H∞Reduction Level γminDerivation algorithm,
Process is as follows:
Step 1: given γ;
Step 2: solvingAndObtain one group of feasible solution Enable k=0;
Step 3: solving non-linear minimisation problem:
WithIt enablesKk=K, Lk=L;
Step 4: whether MATRIX INEQUALITIES solution meets in checking step 4, it is positive real number that satisfaction, which then enables γ=γ-δ, δ,
K=k+1 is enabled, third step is gone to;It is unsatisfactory for and is more than given maximum number of iterations, then exit calculating;
Step 5: showing this optimization problem in given the number of iterations without solution if γ is equal to given value;It is given if γ is less than
Definite value, then γmin=γ+δ;
Step 5:
Select residual error evaluation functionThreshold valueWherein l0For
Initial evaluation moment, L0For evaluation function maximum step-length, pass through comparison JkAnd JthCan be detected out whether faulty generation:
The beneficial effects of the present invention are: firstly, describing sensor respectively to controller using two independent Markov chains
When extend to controller to actuator time delay, network control system is modeled as tool, and there are two the control systems of Markov chain.Herein
On the basis of construct fault Detection Filter and establish closed-loop system model using the method for state augmentation.Then, with matrix
The form of inequality has obtained closed-loop system Stochastic stable and has met H∞Performance condition gives controller and filter gain square
Battle array and H∞The method for solving of Reduction Level, and obtained transition probability and minimum H∞Relationship between Reduction Level.It solves
Network control system under sensor to controller time delay and controller to actuator time delay transition probability all part unknown condition
H∞Fault detection problem.
Specific embodiment
A kind of network control system fault detection method, comprising the following steps:
Step 1:
It utilizesAnd μkSensor in network control system is respectively indicated to the time delay and controller of controller to actuator
Time delay, and respectively in finite aggregateWith N={ μmin, μmaxIn value, transfer is general
Rate matrix is respectively as follows: ∧=[λij], Π=[πrs], λijAnd πrsIt is defined as follows:
In formula:
ForIt enablesWherein,If
ThenWithIt is further represented as Wherein
1≤d≤nM, nMIt is the number for gathering M element;ForIt enables Wherein,WhenThenWithIt is further represented asWherein 1≤g≤nN, nNIt is set N element
Number;
Step 2:
To the network control system with time delay:
In formula: xk∈RnIt is system mode vector, yk∈RmIt is system output vector, dk∈RpIt is L2The external disturbance of bounded
Signal, fk∈RqIt is L2The fault-signal of bounded, Ap, Bp, Bd, Bf, CpIt is the permanent matrix of appropriate dimension, K is control to be designed
Device gain matrix;Fault Detection Filter is constructed in controller end:
In formula:It is filter status vector,It is filter output vector;rk∈RqFor residual error, L be to
Filter gain matrix is designed, V is residual error gain matrix;
Step 3:
Definition status evaluated error and residual error error signalrek=rk-fk;Define augmentation vector Obtain closed-loop system equation:
In above formula:
I1=[0 I]T∈R2n×n, I2=[I 0]T∈Rn×2n, I3=[0 I]T∈R(p×q);
For unknown parameter K and L in processing system, augmentation vector is defined:Then
Closed-loop system is written as:
In formula:
WhereinThePiecemeal is unit battle array, remaining is zero gust,ThePiecemeal is unit
Battle array, remaining is zero gust;
Step 4:
K and L unknown in step 3 in order to obtain are solved:
In above-mentioned formula:
Provide controller gain matrix K and filter gain matrix L and minimum H∞Reduction Level γminDerivation algorithm,
Process is as follows:
Step 1: given γ;
Step 2: solvingAndObtain one group of feasible solution:Enable k=0;
Step 3: solving non-linear minimisation problem:
WithIt enablesKk=K, Lk=L;
Step 4: whether MATRIX INEQUALITIES solution meets in checking step 4, it is positive real number that satisfaction, which then enables γ=γ-δ, δ,
K=k+1 is enabled, third step is gone to;It is unsatisfactory for and is more than given maximum number of iterations, then exit calculating;
Step 5: showing this optimization problem in given the number of iterations without solution if γ is equal to given value;It is given if γ is less than
Definite value, then γmin=γ+δ;
Step 5:
Select residual error evaluation functionThreshold valueWherein l0For
Initial evaluation moment, L0For evaluation function maximum step-length, pass through comparison JkAnd JthCan be detected out whether faulty generation:
The main object of the present invention beAnd μkTransition probability part it is unknown under conditions of, design error failure detection filter
Wave device simultaneously meets:
1) in ωkWhen=0, system Stochastic stable.
2) under zero initial condition, residual error error signal rekMeet following H∞Performance indicator:
Wherein γ > 0 is H∞Reduction Level.
The lemma that processing array inequality is used in the method for the present invention is introduced first:
1, it is provided in existing technological achievements, when given scalar lambdai>=0 and matrix Pi>=0, i=1,2 ..., N, have it is following at
It is vertical:
2, we provide lemma 2 on the basis of lemma 1: for given scalar lambdai>=0, πr>=0 and matrix Pi,r>=0, i
=1,2 ..., N1, r=1,2 ..., N2;ThenPerseverance is set up, which can
To prove to set up by mathematical induction:
Work as N1=1, N2When=1, there is λ1π1P1,1≤λ1π1P1,1, it is assumed that N1=k1, N2=k2, i.e.,
Work as N1=k1+ 1, N2=k2When+1,
By lemma 1, it is known that:
Therefore,
Work as N1=k1+ 1, N2=k2When+1: I.e. lemma must be demonstrate,proved.
Main Conclusions:
Theorem 1 works as ωkWhen=0, closed-loop system:
It is Stochastic stable, and if only if
There are matrix K, L and positive definite matrix Qi,r> 0, Qj,s> 0 makes following MATRIX INEQUALITIES for all i, and j ∈ M, r, s ∈ N is equal
It sets up:
The proof of sufficiency of the theorem:
Choose Lyapunov functionSo,
If the formula is set up:
The minimum that wherein α is-Ξ is special
Value indicative.For any T >=1, as available from the above equation:
Adequacy must be demonstrate,proved.
Necessity proves:
Assuming that systemIt is Stochastic stable, i.e.,It enables:WhereinAssuming that εk≠
0, by above formula andOrthotropicity to be known as below the limit be existing:
By εkArbitrariness can further obtain:
Known to
Consider formula:
As T → ∞, by above formula andΞ < 0 can be obtained, theorem must be demonstrate,proved.
Inference 1 works as ωkWhen ≠ 0, residual error weight matrix V and scalar γ > 0 are given, if it exists matrix K, L and positive definite matrix
Qi,r> 0, Pj,s> 0, Qj,s> 0, so thatPj,sQj,s=I,
In formula:
For all i, j ∈ M, r, s ∈ N is set up, then with the unknown closed-loop system in transition probability part be with
Machine is stable, and meets H∞Performance indicator.
It proves, can be obtained by lemma 2:
Therefore:
In formula:
IfIt sets up, then
For above formula, k is added up from 0 to ∞ to be obtained:
It can be obtained by system Stochastic stable and zero initial condition:
ByIt enablesIt can obtain unknown in step 3 for calculating in step 4
K and L formula:
Ginseng in order to illustrate the validity of the method for the present invention, in Fig. 1 in the network control system state equation of controlled device
Number are as follows:
Assuming thatAnd μkSensor in network control system is respectively indicated to the time delay and controller of controller to actuator
Time delay, and the value in finite aggregate M={ 0,1 } and N={ 0,1 } respectively, in order to obtain transition probability and system performance it
Between relationship, consider following four in situation time delay transition probability matrix:
(1)(2)
(3)(4)
Given residual error weight V=[0.1 0.1], acquires minimum H according to inference 1∞Reduction Level γminIt is as shown in the table:
γ under the different transition probability matrixs of table 1minValue
Transition probability matrix |
A1Π1 |
A2Π2 |
A3Π3 |
A4Π4 |
γmin |
1.0677 |
1.0536 |
1.0493 |
1.0198 |
From table 1 it follows that the information of known time delay transition probability matrix is more, then H∞Reduction Level γminJust
Smaller, i.e. the Disturbance Rejection ability of system is stronger.
Assuming that the original state x of system-1=x-2=[0 0]T, x0=[- 1 1]T, dkFor the random signal for being evenly distributed on [- 0.02 0.02], fault-signal are as follows:The initial mode of network delay And μkRespectively as shown in Figures 2 and 3, for tool
There is transition probability matrix A2And Π2NCS, controller gain matrix K is acquired according to step 4 and filter gain L is as follows:
Choose residual error evaluation function:Acquire failure determination threshold value:
Residual signals, residual error evaluation function and threshold curve are respectively such as
Shown in Fig. 4 and Fig. 5.It will be seen that when an error occurs, the residual error evaluation function of residual signals has from Fig. 4 and Fig. 5
Significant change.In addition, acquiring J11=0.0019 < Jth=0.0066 < J12=0.0234, it means that the 3rd occurred in failure
Time cycle, fault filter just detected failure.
Under conditions of the present invention extends to C-A time delay transition probability in the S-C of NCS part is unknown, the H of NCS is had studied∞
Fault detection problem.New lemma has been obtained to handle unknown time delay transition probability, further in the form of MATRIX INEQUALITIES
Adequate condition existing for fault Detection Filter is given, and gives the solution of filter gain matrix using the thought of CCL
Method.Case Simulation illustrate obtained fault Detection Filter not only to Fault-Sensitive, but also to external disturbance have robust
Property.