Diagnostics of Regular Performance of Water Distribution Networks and its Relation to the Network Reliability - Pattern Recognition for Reliability Assessment of Water Distribution Networks - page 188

The results for simulation runs of the ten networks, keeping the same initial parameters as in
the previous calculations, are given in Table 7.4.

Table 7.4 Pearson correlation of network pipe properties to the loss of demand, 1 - ADF j

No.Name V j Q j v j h f,j S j P j T j pb j pb j /L j hl j PBI j

1 sn73 0.45 0.96 0.96 0.87 0.96 0.77 -0.58 0.50 0.30 0.77 -0.80

2 sn80 0.39 0.83 0.83 0.99 0.98 1.00 -0.08 0.19 -0.17 0.98 -1.00

3 O20sn73 0.91 1.00 0.72 0.18 0.08 0.85 -0.36 0.66 0.67 0.16 -0.01

4 O20sn80 0.93 0.98 0.53 0.00 -0.02 0.87 -0.16 0.29 0.34 0.04 0.07

5 sn0179 0.26 0.85 0.85 0.97 0.97 0.95 -0.30 0.29 0.14 0.96 -0.96

6 sn0199 0.39 0.84 0.84 0.99 0.98 0.98 -0.10 0.31 0.05 0.93 -0.98

7 sn0109 0.02 0.90 0.90 0.95 0.99 0.96 -0.09 0.10 0.39 0.88 -0.90

8 ngt01-50 -0.20 0.80 0.80 0.93 0.96 0.99 -0.07 -0.25 -0.32 0.87 -0.98

9 ngt01-151 -0.10 0.80 0.80 0.95 0.97 0.98 -0.02 -0.16 -0.28 0.89 -0.94

10 ngt01-200 -0.11 0.66 0.66 0.83 0.87 0.95 -0.03 -0.12 -0.08 0.73 -0.91

Although the table shows inconsistent results, a few interesting conclusions can be made:
- There is a clear distinction in the correlation between the GA-optimised networks,
O20snXX, and other oversized networks. In general, the correlation between the loss of
demand and pipe parameters which involve friction loss is fairly good in oversized nets,
whilst it is very bad in the two optimised nets. Oppositely, the optimised nets show
reasonable correlation of pipe volumes with the demand loss, whilst this correlation is
weak in oversized networks.
- The weaker correlation of flows in oversized nets is not a surprise; those have been used
to assess the buffer through NBI . Hypothetically, the weaker the correlation between the
flows and loss of demand could mean the more reliable network. The lowest value of the
Person's index of 0.66 for the biggest network, ngt01-200 , is signalling this.
- The most consistent correlations across the entire range of networks are for the power
loss, P j , which are pretty good for oversized networks and moderately good for optimised
ones. Furthermore, the correlations of the pipe residence time, T j , are consistently bad,
hinting this parameter as inappropriate indicator of potential loss of demand in case of the
pipe failure.

To check the consistency of results for optimised networks, all 30 SN -nets analysed in
Chapter 6 have been optimised and the correlation of the same parameters done; the results
are shown in Table 7.5. Eventually, the same trend has been observed for all the optimised
networks as concluded based on the analysis of the results in Table 7.4. The best correlation
is made with pipe flows, the reasonable one with the volumes and the moderate one with the
power loss. All other parameters show poor correlation.

To get some insight into the pattern of correlation in case of the demand increase, the two
optimised networks from Table 7.4 have been tested for the total demand increase of 32%
( O20sn73 ) and 51% ( O20sn80 ). The results of correlations are given in Tables 7.6 and 7.7,
respectively. No matter how good or bed the correlations are, the trend of improvement
following the growth of demand is visible in both tables. This of course comes with the fact
that both networks are becoming increasingly unreliable with the growth of demand.