Not able to solve the Objective Function for MPC #1230
Replies: 9 comments 26 replies
-
|
Adding that term to the objective makes the problem brutaly much harder, as it leads to a nonconvex problem. Not only is that a massive issue, the code should not even work, as you use the SOCP-represented norm operator in this nonconvex way, so it should simply yield an error message. I cannot test that though as you code does not run (and is completely broken as you haven't placed it in a code block but simply pasted as text) but I guess you haven't checked that it actually solves the problem (you are not looking at any diagnostics in the code) but then kept running the code anyway thus crashing Had you used a 1-norm or inf-norm, the term would be accepted, and lead to a MILP-representation |
Beta Was this translation helpful? Give feedback.
-
|
As far the literature regarding the flight formation using MPC is available I am seeing, I am finding the only possible way was to modify the objective function. Since I am new to the forum I am not very aware of the code cell. `clc B=[0 0 0 x02=[-50, 0,0, 0 0 0]'/1; x03=[-30,0,0, 0 0 0]'/1; xd1=[-30,-20,0,0,0,0]'/1; xd2=[30,-25,0,0,0,0]'/1; xd3=[0,32,0,0,0,0]'/1; %% MPC Parameters nx = 6; % Number of states % MPC data %% x_m(:,1)=x01; for j=1:500 j; % U = sdpvar (3, 1) ; |
Beta Was this translation helpful? Give feedback.
-
|
As I said, you have to put your code in a codeblock (<>). If you look at the code you pasted it is broken |
Beta Was this translation helpful? Give feedback.
-
|
< clc B=[0 0 0 x02=[-50, 0,0, 0 0 0]'/1; x03=[-30,0,0, 0 0 0]'/1; xd1=[-30,-20,0,0,0,0]'/1; xd2=[30,-25,0,0,0,0]'/1; xd3=[0,32,0,0,0,0]'/1; %% MPC Parameters nx = 6; % Number of states % MPC data for k = 1:N |
Beta Was this translation helpful? Give feedback.
-
|
You have to use the code block icon shown in the editor |
Beta Was this translation helpful? Give feedback.
-
|
`clc B=[0 0 0 x02=[-50, 0,0, 0 0 0]'/1; x03=[-30,0,0, 0 0 0]'/1; xd1=[-30,-20,0,0,0,0]'/1; xd2=[30,-25,0,0,0,0]'/1; xd3=[0,32,0,0,0,0]'/1; %% MPC Parameters nx = 6; % Number of states % MPC data for k = 1:N |
Beta Was this translation helpful? Give feedback.
-
|
Use the preview before you post to make sure it actually looks like a code block
|
Beta Was this translation helpful? Give feedback.
-
|
The moment I add, abs or inf norm. The optimizer is not giving any output. The control U is coming out as NAN |
Beta Was this translation helpful? Give feedback.
-
|
Copy the complete code so it can run. As I said, |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
Sir, is it possible if I mail you the code. |
Beta Was this translation helpful? Give feedback.
-
|
No, I don't see what is so hard. Copy the code to the editor. Mark and press code-block. |
Beta Was this translation helpful? Give feedback.
-
|
Sir, it's a matlab code. Hence whenever I am trying to keep it asking for formatting. |
Beta Was this translation helpful? Give feedback.
-
|
You obviously managed to do it above, as you copied and correctly marked as code, which I copied to my matlab, and now copied back here |
Beta Was this translation helpful? Give feedback.
-
|
Ok, sir. Got it. Will do and send you right away. Thanks |
Beta Was this translation helpful? Give feedback.
-
|
@johanlofberg sir, is it possible if I mail you the code. |
Beta Was this translation helpful? Give feedback.
-
|
This is working code. |
Beta Was this translation helpful? Give feedback.
-
|
Ok sir. Thanks. Hopefully will get back with some positive results. |
Beta Was this translation helpful? Give feedback.
-
|
Hi, nx = 6; % Number of states but it is not happening |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
Sir, following the basic of Yalmip for implicit. I am trying to convert the code for state space. With Input as 6 States and Control Sate. But i AM NOT ABLE TO UNDERSTAND THE architecture of the
|
Beta Was this translation helpful? Give feedback.
-
|
There is no question here. Also, when you show code, put it in a single code block, it looks horrible at the moment. |
Beta Was this translation helpful? Give feedback.
-
|
Sir, my System is The changes I made from the original % MPC data The other changes were my own dynamics matrices. |
Beta Was this translation helpful? Give feedback.
-
|
Unless you mean you simply want to look at the computed value in your iteration. Then you define the objective as an output in the optimizer object |
Beta Was this translation helpful? Give feedback.
-
|
Sir, |
Beta Was this translation helpful? Give feedback.
-
|
hence you output both inputs and objective, |
Beta Was this translation helpful? Give feedback.
-
|
Sir, can you pleas tell why using 1 in norm function is giving answer while if I am removing 1 there is no solution? |
Beta Was this translation helpful? Give feedback.
-
|
not sure what you mean with no solution. If you remove the 1, you will get a 2-norm, and since you use gurobi it should be capable to solve the problem As always, when developing code where you haven't resolved all issues and is confident everything works you should never use an optimizer as that adds complications, and if you absolutely must use optimizer you should set verbose to 2 so you see what the solver is doing. optimizer constructs are for speeding up simulations only. to make matters worse, you are not looking at the data you compute, or the diagnostic flag, and thus miss the fact that you start sending matrices to the optimizer object instead of a vector. this is coming from the fact that the model has three inputs, but you extract U(1) (and this size confusion is visible already in the code as you define R to be a scalar) |
Beta Was this translation helpful? Give feedback.
-
I am working on a problem of satellite formation. I am able to generate profile using YALMIP and Gurobi by minimizing the following function.
objective = objective+norm(x-xd1)+ 0norm(Qx,1) + 1norm(Ru{k},1);
where xd are the final reference states.
but let us suppose i want to make keep a certain distance between the satellite and the reference pints, lets say 10 meters, the code is not giving any output.
objective = objective+abs(norm(x-xd1)-10)+ 0norm(Qx,1) + 1norm(Ru{k},1);
full code
clc
clear
n0=0.00115691;
A=[0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
3n0^2 0 0 0 2n0 0
0 0 0 -2*n0 0 0
0 0 -(n0)^2 0 0 0];
B=[0 0 0
0 0 0
0 0 0
1 0 0
0 1 0
0 0 1]/1000;
C=eye(6);
D=0;
Sys1=ss(A,B,C,D);
Dis_mod=c2d(Sys1,0.1);
Ad=Dis_mod.A;
Bd=Dis_mod.B;
Cd=Dis_mod.C;
Dd=Dis_mod.D;
rank(ctrb(Ad,Bd));
%%
x01=[-70, 0,0, 0 0 0]'/1;
x02=[-50, 0,0, 0 0 0]'/1;
x03=[-30,0,0, 0 0 0]'/1;
xd1=[-30,-20,0,0,0,0]'/1;
xd2=[30,-25,0,0,0,0]'/1;
xd3=[0,32,0,0,0,0]'/1;
%% MPC Parameters
nx = 6; % Number of states
nu = 3; % Number of inputs
% MPC data
Q = eye(6);
R = 1;
N = 2;
%%
x_m(:,1)=x01;
for j=1:500
x = x_m(:,j);
u = sdpvar(repmat(nu,1,N),repmat(1,1,N));
constraints = [];
objective = 0;
for k = 1:N
x = Adx + Bdu{k};
objective = objective+norm(x-xd1)+ 0norm(Qx,1) + 1norm(Ru{k},1);
constraints = [constraints, -100 <= u{k}<= 100];
end
%opts=sdpsettings('solver','gurobi','OutputFlag',0);
%opts = sdpsettings('solver','sdpa');
%optimize(constraints,objective,opts);
optimize(constraints,objective);
U_IN(:,j)=value(u{1});
x_m(:,j+1) = Adx_m(:,j) + BdU_IN(:,j);
clear constraints;
clear u;
end
figure(1)
plot(x_m(1,:),x_m(2,:))
%%
Beta Was this translation helpful? Give feedback.
All reactions