suboptimal mpc

[shariat]

Dear All

I want to use YALMIP to solve sub-optimal MPC problem. My optimization problem is to minimize a finite horizon quadratic cost of the state and control trajectory, while satisfying constraints. On the other hand, I want to specified maximum number of iteration and supply initial guesses to solvers. Is it possible with a solver such as CPLEX Or not?

 

Best regards

[Johan Löfberg]

you would have to search through all the available options you have in the cplex options structure (or read the cplex API reference)

[shariat]

Dear Prof. Johan Löfberg

Thanks for your helpful guide.

I wrote a routine for testing optimal and suboptimal MPC (  attached file ). In suboptimal MPC, the number of iterations is limited and initial solution is selected by warm-start. ; It seems that suboptimal MPC take longer time than optimal MPC.

Surprisingly, it seems that suboptimal MPC  have a longer execution time than optimal MPC time. Please comment about this issue.

Thanks

Shariat

[Johan Löfberg]

If the solver implements a primal-dual solver, perhaps with yet more varibles to represent slacks, the fact that you can supply a (hopefully) relevant primal (relative to the final solution) does not necessarily help the solver, as that solution lacks dual information forcing the solver to start in a region which is bad from the solver algorithm perpective (far away from central path etc). Warm-starting interior-point solvers is hard, and almost never gives any performance benefits unless much more effort is devoted to the strategi