Problem solved and it was completely my fault.  I was using the load2disp 
function in such a way that it was causing the network to grow each dispatch 
period.  I've changed how I use it and the computation time for each period 
tracks (inversely) with the load for that period.  The unit de-commitment also 
now appears to run correctly each dispatch period, evaluating each period based 
on the load for that period.

Thanks for the guidance and help.

Trevor

From: [email protected] 
[mailto:[email protected]] On Behalf Of Shri
Sent: Tuesday, July 02, 2013 11:04 AM
To: MATPOWER discussion forum
Subject: Re: Smartmarket uopf computation time

Matlab code profiler might be helpful to figure out which part(s) of the code 
are the most time consuming.

Shri
On Jul 2, 2013, at 10:43 AM, Hardy, Trevor wrote:


It is only during the first period or two that I see messages in the Matlab 
console indicating that generators are shutting down.  For the vast majority of 
the periods, these messages are not displayed but the computation time 
continues to grow. It looks like I'll need to dig into this.


Trevor

From: 
[email protected]<mailto:[email protected]>
 [mailto:[email protected]]On Behalf Of Ray Zimmerman
Sent: Tuesday, July 02, 2013 10:33 AM
To: MATPOWER discussion forum
Subject: Re: Smartmarket uopf computation time

I expect that the computation time might grow quadratically with the number of 
units it needs to shut down. So if your simulation has a pattern of decreasing 
loads you might see this trend. Otherwise, I'd be suspicious that something 
else is going on.

Also, I'm sure you are aware, but just as a reminder, the algorithm implemented 
here is only a heuristic and a pretty computationally expensive one at that. It 
is not suitable for large systems with lots of generators that need to be shut 
down. For such a case, it might make more sense to run it with all generators 
on with PMIN set to zero, then use generators with non-zero dispatches as a 
starting commitment for MATPOWER's current algorithm.

--
Ray Zimmerman
Senior Research Associate
B30 Warren Hall, Cornell University, Ithaca, NY 14853
phone: (607) 255-9645






On Jul 2, 2013, at 10:47 AM, "Hardy, Trevor" 
<[email protected]<mailto:[email protected]>> wrote:



Dr. Zimmerman and company,

First of all, thank you for your efforts on MATPOWER and the continued support 
you provide.  It has been a wonderful tool to use.

I'm using the smartmarket module on a 240 bus system over many one-hour 
dispatch periods.  I recently ran this system for one week of simulation time 
(168 dispatch periods) and noticed that the computation time per period grows 
quadratically as the simulation progresses.  I've read the section of the 
manual on the uopf function (called by the smartmarket function) and it is not 
clear to me if this is expected behavior.  If it is, is there anyway to 
mitigate the growing computation time?

Trevor Hardy
Wichita State University

Reply via email to