In my experience, relative performance between different solvers (e.g. MIPS,
IPOPT, Knitro) can vary widely from problem to problem, so the solvers that are
faster on average, may not be faster on a particular problem. On my system,
using IPOPT with PARDISO, I get the following solve times and numbers of
MIPS: 16 sec, 45 iterations
IPOPT: 22 sec, 124 iterations
Knitro: 38 sec, 117 iterations
Unfortunately, the page at http://pardiso-project.org only includes pre-built
MEX binaries for Linux and Mac, not Windows.
> On Aug 31, 2016, at 11:51 PM, winsonwoods <winsonwo...@163.com> wrote:
> Dear Matpower Community,
> I am a Ph.D student of SWJTU from China. When I tried the
> matpower6.01b1 to solve the CASE9421PEGASE, I found a very strange things.
> When I use the default MIPS solver, it just need about 22 seconds; when I
> change to the IPOPT solver (I have installed the OPTI toolbox), it need about
> 63 seconds. In my expectation, the IPOPT should be faster than the MIPS as
> the matpower6.01b1 manual said. I guess the reason is that the IPOPT should
> be used with PARDISO to improve performance.
> However, I can't find the Pre-built MEX binaries of high-performance
> IPOPT-PARDISO for the WIN-64bit platform and, I never successfully compile
> the PARDISO matlab interface. Does anyone have compile successfully? Could
> you share the code or the precompile mex file with me?
> Yours Sincerely,
> Winson Woods.