Environmental Engineering Reference
In-Depth Information
1817 layouts satisfying the configuration of the network template. Those have been then
optimised and analysed on various reliability measures against the costs comprising the
network investment, operation and maintenance. The least of conclusions from the
programme performance and the results of the case study can be summarised in the following
bullets:
-
NEDRA has proven to be pretty robust package that can process a very large sample of
networks within relatively short period of time. The bottle-neck time-wise is in the
network generation module; not necessarily because this process is slow, but mostly due
to huge number of networks that is to be generated to arrive at sufficient variety of the
layouts. To produce thousands of layouts of moderate size, several hours will normally be
needed. That can be seen as a long time, but the whole network generation is one-time
activity, anyway. Running, say, a 48-hour calculation on a fast computer to arrive at e.g.
50-100 thousand layouts is not unthinkable assignment.
-
Single design of a network would rarely deal with thousands or even hundreds of
nodes/pipes. Yet, the number of combinations in even a moderate network size of a few
dozen nodes can be enormously high, both in random and non-random generation
process. In this generation approach, the large number of combinations will be regularly a
must in order to arrive at sufficiently versatile layouts. This variety and the shorter time of
generation can also be enhanced by careful selection of the network generation settings.
-
The network filtering process has been very quick, but the number of layouts that comply
with the network template can be surprisingly low. This will depend on the number of
generated networks and complexity of the template. In general, the more complex the
template and the larger number of generated layouts, the larger number of filtered layouts
will be. To further increase this number, it may be sensible to also filter the layouts that
have a few connections not matching the template, because in an otherwise promising
layout, these links can be manually reconnected in the final stage.
-
The selected EO-optimiser has proven to be pretty fast, but not arriving at quite logical
sequence of diameters, typical for engineering practice. For instance, it has been not that
uncommon to see a series of pipes where larger diameter pipe is inserted in between the
two pipes of smaller diameters. Yet, even in this kind of layout, the optimiser has
managed to find combinations of fairly satisfactory reliability because quite some layouts
were optimised creating a buffer, which somehow defeats the assumption elaborated in
previous chapters that GA makes distinctly branched skeleton of larger pipes within
looped configurations. This phenomenon should therefore be investigated further.
Secondly, the EO-optimiser has found some solutions resulting in pressures below the
preferred threshold. Thorough assessment/selection of GA-optimiser and input settings
has not been within the scope of this research and it is yet to be investigated to which
extent the anomalies in the results can be eliminated by more appropriate selection of the
settings, in the first place the number of generations and the penalty. All the GA-
simulations have been run only once per batch of networks and with fixed settings; hence,
no consistency could have been found to point the source of peculiar results. What could
however have been observed is that the pressures below the threshold were more common
for some classes of generated networks, e.g. those of lower complexity. Eventually, those
have not qualified anyway, for the sake of their low connectivity i.e. low reliability at
relatively high price. Reasonably looped layouts, such as those finally analysed, were
optimised with the minimum pressure that has fallen regularly around the threshold of 20
mwc.
-
Although not entirely justified, a (properly) GA-optimised design appears to be good
starting point in reliability analysis; it avoids any bias possible by visual assessment of
network configurations, which can be easily deceiving. GA provides quick reference
Search WWH ::
Custom Search