I have not really used MATPOWER's OPF for distribution systems, but here are my 
thoughts on a number of the solvers for transmission networks.

MINOPF - very fast and very reliable on small systems ~ 100 buses or less 
(limited to 32-bit Matlab versions)
MIPS - reasonably fast even on very large systems, not always reliable (pure 
Matlab, no MEX)
fmincon - usually slower than MIPS, but sometimes more reliable
TSPOPF - same as MIPS, very slightly faster
IPOPT - similar to MIPS, probably more reliable, occasionally faster on very 
large systems (1000's of buses)
KNITRO - very fast and robust, including on large models
SCPDIPM, MIPS-sc - similar to TSPOPF and MIPS, but will sometimes solve 
problems not solved by them
TRALM - slower and not necessarily and probably less reliable than MIPS, 
doesn't scale as well as interior point solvers

So, it depends what you have available. MIPS is always available and is usually 
a pretty good choice unless you are running into numerical reliability 
problems. While I have less experience with it, KNITRO seems to be a very good 
solver.

If you have multiple options, it is often a good idea to simply test the 
various solvers on your problem. I typically use MINOPF for the best results on 
small problems (~100 buses or less) and MIPS or TSPOPF on larger problems 
unless they have reliability problems, in which case I try out other options.

-- 
Ray Zimmerman
Senior Research Associate
419A Warren Hall, Cornell University, Ithaca, NY 14853
phone: (607) 255-9645




On Feb 13, 2012, at 12:12 PM, Roberto Carvalini wrote:

> Dear Ray,
> 
> For a large distribution system such as 113 bus test system, in order to do 
> an OPF or a runmarket, could you please let us know which solvers are better, 
> i.e. KNITRO, SCPDIPM, MIPS?
> 
> Regards
> 
> Roberto 
> 

Reply via email to