uncertainties in the objective

[Yangyang Li]

Hi Johan,

I'm a starter of YALMIP, and I'm trying to set up a robust LP model, which has uncertainty variables in the objective with YALMIP. I used CPLEX as the solver. 

The problem is for example, x, and y are the certain variables, z are the uncertain variable, and I want to set z as ellipsoidal uncertainty, with, for example, (2, 3) as the center. I used norm to realize this uncertain model: 

What I used is like this:

x = sdpvar[100,1];

y = sdpvar200,1];

z= sdpvar[2,1];

obj = z(1)*sum(x) + z(2)*sum(y); 

constraints = [uncertain(z), norm(z-[2;3],2) <= 1];

constraints = [constraints, ....]

solvedsp(constraints, obj);

I used double(x), double(o) for checking the values of the certainty variables, and I get them. But when I checked the value of the uncertain variable z with double(z), I got [NaN,NaN]. Does YALMIP has a specific command for save the values or statistics for uncertain variables? And how can I save the obective value after solving the problem?  Did my method to set up ellipsoidal uncertainty within the objective correct?

Thanks a lot.

[Johan Löfberg]

Worst-case robust optimization is typically based on elimination of the uncertainties and derivation of an equivalent problem without uncertainties, i.e., the uncertain variables are never part of the optimization problem. Hence, the actual value which constitutes the worst case is never computed.

Your model looks reasonable (except for a typo and the fact that you use brackets when defining the variables)

[Yangyang Li]

Thank you very much for you help, Johan.