State Space with Disturbance

Question

Azaldo Machava 2021-1-21

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/723093-state-space-with-disturbance

编辑： Paul 2025-4-28

I have this system:

Xdot = AX +Bu +Fd,

where A,B and F are the system, input distribution and disturbance distribution matrices respectively and X, U and d are the state, control and disturbance vectors respectively.

i have all the matrices data. i want to know how to use matlab to plot the state variables for a step disturbance.

where i am finding problem is because the standard form of state space in matlab is :Xdot = AX +Bu without the distubance vector.

so how do i input my system in Matlab???

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Paul 2021-1-21

4
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/723093-state-space-with-disturbance#answer_603218

编辑：Paul 2024-12-15

在 MATLAB Online 中打开

Combine d and u into a single input vector:

Xdot = A*X + [B F] * [u;d]

y = C*X + [D 0]*[u;d] % assuming the disurbance doesn't feed directly into output

sys = ss(A,[B F],C,[D zeros(size(C,1),size(F,2))])

24 个评论
显示 22更早的评论隐藏 22更早的评论

John 2023-2-24

编辑：John 2023-2-24

@Paul Just to check, does your augmenting vectors (where [u;d] are the inputs) allow independent control of u and d using lsim()? I ask, because I tried the below, and am seeing an unexpected (to me) result where an input disturbance of 1 does not give a position shift of 1.

For example, assume some basic 2nd order system for the open-loop SS model (m = 1e-5, k = 62534, b = 0.0016)

openLoopSsSys = ss(A, [B F], C, D)

with A = [0 1; -6.3e4 -160], B = [0; 6.3e4]

x1 = position

x2 = velocity

Assume the lsim() time is 0 to 1 seconds, and the input parameter to lsim() is [u(t); d(t)] , instead of just the standard u(t), as you proposed:

lsim(openLoopSsSys, [u; d]).

Matlab seems to correctly simulate a closed loop response to a reference input, and closed-loop response to a disturbance, if the input vector is defined as (for example) below

u = 1 in t = [0, 1] (Ie all the simulation time)

d = -1 if t > 0.75, d = 0 otherwise;

I do see some change, but not as expected.

When I added d = -1 as above, there was virtually no x1 change output. The "d" input must be much different -1 (-400), even though the mapping above should be:

x1_dot (position change) = x2 + d + u

I thought an input d = -1 would shift x1 (position) by -1, instead of needing to be (in this example) -400.

Why is that? Is lsim() somehow interpreting this differently than i thought, or is my understanding incomplete on how d maps to x1?

Thanks.

Paul 2023-2-24

编辑：Paul 2023-2-24

在 MATLAB Online 中打开

Hi John,

Looks like we were both typing at the same time. Trying to reconstruct the example

m = 1e-5; k = 62534; b = 0.0016;
h = minreal(tf(1,[m b k]))  % tf from force to position
h =
 
           1e05
  ----------------------
  s^2 + 160 s + 6.253e09
 
Continuous-time transfer function.
sys = ss([0 1;-k/m -b/m],[0;1/m],[1 0],0)
sys =
 
  A = 
               x1          x2
   x1           0           1
   x2  -6.253e+09        -160
 
  B = 
          u1
   x1      0
   x2  1e+05
 
  C = 
       x1  x2
   y1   1   0
 
  D = 
       u1
   y1   0
 
Continuous-time state-space model.
tf(sys)  % verification
ans =
 
           1e05
  ----------------------
  s^2 + 160 s + 6.253e09
 
Continuous-time transfer function.

That doesn't match your system, the B-matrix is different. Maybe your model is different than what I assumed?

Also, based on the model posted

format short e
A = [0 1; -6.3e9 -160], B = [0; 6.3e4]
A = 2×2
1.0e+00 *

            0   1.0000e+00
  -6.3000e+09  -1.6000e+02
B = 2×1
           0
       63000

The equation for x1dot is (after adding the disturbance into the problem): x1dot = x2 + 6.3e4 * u - d

So u will have a much, much bigger effect on x1dot than d.

If you'd like to continue the discussion, I suggest posting a complete code, show the results from running that code, and describe what about those results are of concern.

John 2023-2-25

编辑：John 2023-2-25

matlabPost.m

Thanks @Paul. You're right, that was confusing. Let me start over and be more specific with another related question, that's more fundamental to my questions. As you requested a test file is attached.

Here's an open-loop 2nd order system with a disturbance vector added per your recommendation. Since it's a basic 2nd-order spring-mass-damper system, and the state space was set up in the standard way, the output states are x1=position, x2=velocity. I'm plotting the lsim() output states x in response to a constant reference step at t=0 (the sys is normalized), and dist step at t = 0.25.

When plotting all state variables, the position changes as expected, and the velocity decays to 0 as expected after settling after a reference command.

But the velocity stays constant after a disturbance (force) input. Presumable, since the position stabilizes (forces in equilibrium), velocity should be 0.

As Matlab treating this input vector correctly?

(The code should generate this plot as well)

Paul 2023-2-25

在 MATLAB Online 中打开

Hi John,

The short answer is that I think the intent was to add in the disturbanc as a force, i.e., at the acceleration level, but it was actually added in at the velocity level.

>> sysOlWithDistDc
sysOlWithDistDc =
 
  A = 
                  x      x dot
   x              0          1
   x dot  -6.25e+04       -150
 
  B = 
            r(ref)   F(dist)
   x             0     416.7
   x dot  6.25e+04         0
 
  C = 
            x  x dot
   pos      1      0
 
  D = 
         r(ref)  F(dist)
   pos        0        0
 
Continuous-time state-space model.

The equation for x1dot is: x1dot = x2 + 416.666666 * Fdist. So, with Fdist = -1 the steady state value of x2 will be 416.6666666, which is exactly what is seen. More generally, for an input of r = 1 and d = -1, we know this system will have an equilibrium point with with states

>> -sysOlWithDistDc.A\(sysOlWithDistDc.B*[1;-1])
ans =
                         0
     4.166666666666666e+02

So the equilibrium position is x1 = 0, also as seen.

If the disturbance was intended to be a force as indicated in the comments, then change this line

%% Add open-loop disturbance
% Disturbance input is a force input (not position)
% F = [1; 0];
F = [0;1]; % force sums into x2dot !

and you'll get a reponse you're expecting with x1 = constant and x2dot = 0 in the steady state.

Also, the code is using a variable called Ctf, which isn't defined, but that is easily fixed.

I'm not quite sure what the intent is of all the dcgain scaling .... but that doesn't matter for this analysis, other than how it affects the steady state values.

John 2023-2-26

编辑：John 2023-2-26

1) Thanks @Paul, I see. But then since x2 = k*F - x1dot, how is Fdist a velocity disturbance input to the original spring/mass/damper system?

It's just connecting a disturbancec only to specific states, and breaks the original SS equation set that describes the physical system.

I'm looking for disturbance inputs that stimulate physical systems that behave like one.

Ie with a velocity input, position should also change (x1=ramp, x2=Const).

With a force input, equilibrium should be the result (x1=Const, x2=0).

So isn't this a non-physical system with the dist spec-d as above? How would I spec a vel dist input that affects the spring/mass/damper?

Or what am I missing? :)

2) How is the Matlab native K disturbance matrix used in sys() relate to this strategy of augmenting a disturbance to SS? K seems a mapping, presumably to take e(t) to the appropriate state variable (or combinations).

"sys = idss(A,B,C,D,K) specifies a disturbance matrix K.

"A state-space model of a system with input vector u, output vector y, and disturbance e takes the following form in continuous time:

dx(t)dty(t) = Ax(t)+Bu(t)+Ke(t)"

(a) how does K support multiple inputs (eg ref and dist)

(b) how is K used concretely? The documentation has no example for an input e(t).

Is K even suggested for use? :D

John 2023-2-27

编辑：John 2023-2-27

Thanks @Paul; I appreciate your help. I'll try to be more clear; apologies for being confusing:)

"At this point, I'm not sure what you're trying to do."

I'll try to restate the question more clearly, and ask a concrete question at the end.

I'm modeling a basic open-loop 2nd order spring-mass-damper system with a disturbance vector input, to model disturbances (F augmentation per your recommendation). The state space was set up in the standard way as in my post above; so the output states are x1=position, x2=velocity=x1dot. The physical motion as a result of the disturbance inputs, must always be representative of how this physical system would actually behave.

Using ss() and lsim(), I'd like to both model

a) a force F to create a position disturbance, eg a step F, and

b) a velocity disturbance -- needs a changing F, I think -- eg a nonzero vel in steady state (after system has been accelerated). It could be an infinitely accelerating frame so the dF is constant and leads to constant velocity (and thus ramping position change). Or it could be eg a shaking table the system's reference frame is attached to, so though the disturbance vel mean could be zero still, but steady state is still not zero vel.

I'd like to plot lsim() output states x in response to (a) and (b). Let's start with velocity since that's what the .m file has, and since you mentioned that I already had a "velocity disturbance". (Can deal with F another time). A true velocity disturbance should cause the system's position (x1) to change continuously.

In a physical system, this nonzero vel (x2=x1dot) would only be caused be a changing F I think; a steady-state F would be a position disturbance (or an intentional reference input), and cause equilibrium and x1dot-->0. Per above, this nonzero constant vel could be an accelerating system, for example, or if non-constant vel, it could be a shaking system.

Questions: For

[B, F]; F = [1; 0];

What do you mean by "velocity disturbance" augmentation in the context of this physical system? Do you just mean a disturbance into the measurement? If so, why is it inputted to the state vector?

It doesn't seem like it's what I'm hoping for ("someone disturbs the spring/mass/damper system to cause it to move with some velocity for a long time, ie not a position step change"), since velocity disturbance steps are impossible for a system with mass; only the requested dF is possible.

Best,

John

Paul 2023-2-27

编辑：Paul 2023-2-27

在 MATLAB Online 中打开

Perhaps the problem is that the system that is represented by the state space model is not the system that you actually want to represent. If you want x1 to be the position of the mass relative to a fixed reference frame, and if you want x2 to be the velocity of the mass relative to the same reference frame, the model must have the equation

x1dot = x2

Since that's not the equation in your model, either the model is incorrect, or you want to model a different system. As best I can tell, that's the equation you want in your model.

The second equation should just be x2dot = (sum of forces)/mass, or

x2dot = (-k*x1 - b*x2 + u + d)/m,

where u (what you called rref) is a force input, perhaps one that will ultimately be determined from feedback control, and d is force disturbance that is some function of time applied by Mother Nature. So the second equation is

x2dot = -k/m *x1 - b/m * x2 + u/m + d/m

We can write these equations as

xdot = [0 1; -k/m -b/m] * x + [0;1/m] * u + [0;1/m] * d = A * x + B * u + F *d = A * x + [B F] * [u;d]

If the ouputs are just x1 and x2, then we have

y = I * x = C *x.

Putting in the numbers:

zeta = 0.3;
wn = 250;
m = 1e-5;
b = 2*zeta*wn*m;
k = wn^2*m;
A = [0 1;-k/m -b/m];
B = [0;1/m];
F = [0;1/m];
C = eye(2);
D = zeros(2,2);
sys = ss(A,[B F],C,D,'StateName',{'pos' 'vel'},'InputName',{'u' 'd'},'OutputName',{'pos' 'vel'})
sys =
 
  A = 
              pos        vel
   pos          0          1
   vel  -6.25e+04       -150
 
  B = 
            u      d
   pos      0      0
   vel  1e+05  1e+05
 
  C = 
        pos  vel
   pos    1    0
   vel    0    1
 
  D = 
        u  d
   pos  0  0
   vel  0  0
 
Continuous-time state-space model.

Now simulate the response of the system to constant input u = 1 for all time, and constant disturbance d = 5 starting at t = 0.5

t = (0:1e-3:1).';

u = 0*t + 1;

d = 5*(t > 0.5);

y = lsim(sys,[u d],t);

figure

plot(t,y(:,1)),ylabel('pos')

plot(t,y(:,2)),ylabel('vel')

As expected, we see a lightly damped response, the steady state position is constant, and the steady state velocity is zero.

Now change the disturbance input so that it's a ramp force instead of a step

d = 50*(t-0.5).*(t > 0.5);

y = lsim(sys,[u d],t);

figure;

plot(t,y(:,1)),ylabel('pos')

The position goes to a constant until the ramp force disturbance hits, and then the position ramps up as well.

Plot the velocity and zoom in.

plot(t,y(:,2)),ylabel('vel')

xlim([0.4 1])

As expected, steady state velocity is zero for t < 0.5, and then goes to a constant after the disturbance hits, which should be the slope of the position (after the transient due to the disturbance dies out).

These results are all as expected based on the model and are consistent with expectations based on the physics of the problem.

Now, if we change the state-space realization of sys such that F = [1;0], we are no longer representing the spring/damper system we started with. It's now the state-space model of a different system. That system might not even represent a physical system, but math and Matlab don't care, even if you don't change the StateName or InputName. The math is allowable and Matlab will compute the response for that ss model, whatever that response may be.

BTW, I don't think I ever used the term "veclocity disturbance," at least I didn't mean to. I did say " it was actually added in at the velocity level." Here, "velocity level" just means that the disturbance was added into the x1dot equation, as opposed to adding it to the x2dot equation.

John 2023-2-27

Thanks @Paul.

1) "Perhaps the problem is that the system that is represented by the state space model is not the system that you actually want to represent"

Huh...could you clarify?

I thought the system in the .m file i shared, with x1dot = x2, is the same as your post:

A = [0 1; -k/m -b/m];

B = [0; 1/m];

is from the .m file, and

A = [0 1;-k/m -b/m];

B = [0;1];

is from your code above (I think you might have forgotten to divide by m in B, but that doesn't impact anything beyond scaling).

What's the difference? :D Sorry, i'm guessing it's obvious, but I'm missing it...

2) "BTW, I don't think I ever used the term "veclocity disturbance,"

Ah, quite so -- you're absolutely right. Apologies for misrepresenting!

3) "Now, if we change the state-space realization of sys such that F = [1;0], we are no longer representing the spring/damper system we started with." "The math is allowable and Matlab will compute the response for that ss model, whatever that response may be."

Got it -- that makes sense. I'd overlooked that the SS formulation specifically only makes physical sense for a force input at the acceleration level (F=ma), ie x2dot=x1: level.

Tomo 2025-4-27

编辑：Tomo 2025-4-28

Hello @Paul,

If you model it this way like you said,

Xdot = A*X + [B F] * [u;d]

Does that mean to determine reachability we would compute W = ( B_{new}, AB_{new},...) where B_{new}=[B F].

Thanks.

Paul 2025-4-27

编辑：Paul 2025-4-28

If reachability is to be assessed based only on the control input to the system (as is typical), then the reachability matrix would be computed from A and B.

请先登录，再进行评论。

State Space with Disturbance

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

24 个评论
显示 22更早的评论隐藏 22更早的评论

更多回答（0 个）

另请参阅

类别

标签

产品

Community Treasure Hunt

State Space with Disturbance

0 个评论 显示 -2更早的评论隐藏 -2更早的评论

采纳的回答

24 个评论 显示 22更早的评论隐藏 22更早的评论

更多回答（0 个）

另请参阅

类别

标签

产品

Community Treasure Hunt

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

24 个评论
显示 22更早的评论隐藏 22更早的评论