Dynamic programming

[hassouf ayoub]

Hello everyone,

Does PSOPT uses dynamic programming solver for switched systems?

Best regards.

[Wen-Loong Ma]

Don't think so. It's also not particularly designed for switched systems.

--
You received this message because you are subscribed to the Google Groups "PSOPT users group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to psopt-users-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Wen-Loong Ma]

Oh that. That's a tricky problem only switched system will encounter. If your event function is purely a function a time, one way to go is to specify your time horizon for certain switch surface and run DP for each domain, i.e., running multiple QPs.

Please make sure to reply to the group so more people can join.

On Wed, Feb 6, 2019 at 7:56 AM hassouf ayoub <hassou...@gmail.com> wrote:

Yes anyone of them would be a good idea so that I try to apply them for switched systems. I used some of them like Bocop but the problem is that my switch is related to a certain time all the solvers I used don't give the chance to constrain my switch at a certain time.

Le mer. 6 févr. 2019 à 16:48, Wen-Loong Ma <wenlo...@gmail.com> a écrit :

What kind of DP are they referring to? MPC, descending, ascending, discrete, finite horizon? They are all kinds of QP or different version of QP, like QCQP right? That's why there is not solver to DP, but there are QP LP QCQP solvers 

On Wed, Feb 6, 2019 at 7:45 AM hassouf ayoub <hassou...@gmail.com> wrote:

Any advices?

A solver you might know that at least uses dynamic programming?

[hassouf ayoub]

I thought of doing at first, but since this is my master thesis, my professor asked not to do it. He wants from me more effort so that I learn more. So he told me to look for other solutions

[Wen-Loong Ma]

It is indeed a tricky problem, you probably need to customize the optimizer for different systems. I suggest you use this method to get some initial result, and have a better idea about the attributes of the system, and then go from there. But DP normally linearizes the system, PSOPT does not. DP widely used for online control, but PSOPT is offline optimization.

[hassouf ayoub]

Thanks a lot for your help. I'll try what you told me and see what it gives.