Closed Mathadon closed 6 years ago
Result differences can however be fairly substantial:
These differences are mostly caused by the different discretisation method. When the number of elements is set sufficiently high then both discretisations lead to the same result as illustrated in the figure below. However, for the current discretisation level they don't. So I suppose the question is: do we want to increase the default discretisation level? Doubling it does not seem to be sufficient.
In this plot 'n' is nStaRef
, the number of states for a reference slab of concrete of 20cm if we want to get a certain time constant for the resulting R*C.
What I conclude from this is that the short-term response for 'state in wall' is better, but the long term response is worse. Since we're mostly interested in long term responses 'state on wall' seems preferable.
When looking at an actual model (bui600
) the results are similar:
The plotted variables are TSensor
.
This seems to suggest that 'state on wall' is more accurate for low values of nStaRef
. 'state in wall' is much less favourable in this comparison. Based on the earlier figure this could be explained because in a real building we are essentially always looking at the response of disturbances that happened quite a while ago, so the long term response is more important.
Earlier posts were corrected and edited to avoid spamming everybody with e-mails.
I conclude from this discussion that the current implementation is ok. Feel free to comment if you disagree!
I ran a test (
Bui600
) using:and I compared some statistics.
Master:
Benchmark (https://github.com/open-ideas/IDEAS/tree/issue683_benchmark):
cpu time is 2.4 seconds vs 38 seconds.