[NEMO-devel] Result difference

Ben Elliston bje at air.net.au
Thu Sep 28 12:55:47 AEST 2017

Hi Yusak

First, rather than sending huge amounts of output to the list, can you
consider using a web service like pastebin.com and sending a link
instead? Thanks.

On Thu, Sep 28, 2017 at 09:35:54AM +1000, Yusak Tanoto wrote:

> I just realized that I have experienced results difference between what is
> listed right after the iteration finished (Jamali-PV = [.........]) and the
> following breakdown after running 100 iterration, although the simulation
> satisfy the reliability constraint (unserved energy 0.002 and LOLP 0.274).

> 95      15      51.3926         51.3826
> 96      15      51.3702         51.3702
> 97      15      51.4275         51.3702
> 98      15      51.4273         51.3702
> 99      15      51.6488         51.3702

If I understand correctly, you are asking why the result for iteration
#99 is higher than the final result.  This is because the CMA-ES
algorithm does not guarantee that each generation will always be
better than the previous.  It often explores more widely and can
produce a poorer solution than the overall best.  That is what the
"hallfame" (Hall of Fame) column shows you -- the all-time best. In
this case, that is $51.3702/MWh.

Cheers, Ben
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://lists.ozlabs.org/pipermail/nemo-devel/attachments/20170928/08062e38/attachment.sig>

More information about the nemo-devel mailing list