Automated leak localization performance without detailed demand distribution data

Figures & data

Table 1. Leak localization performance (%FP nodes) of the artificial leaks during night, morning peak and daytime and the performance when the correlations are accumulated (accumulated C) for the real leak.

Time	Location	Leakage size(m^3/h)	Qleak/Qtotal	Uniform demand(FP nodes)	Factorized demand(FP nodes)
Night	I	7.5	0.64	74.4%	72.2%
		15	0.72	41.7%	41.1%
	II	7.5	0.78	2.5%	4.0%
		15	0.96	1.0%	1.2%
	III	7.5	0.82	1.0%	1.3%
		15	0.91	0.8%	1.7%
Morning peak	I	7.5	0.13	14.7%	5.0%
		15	0.27	41.0%	14.0%
	III	7.5	0.17	2.2%	0.8%
		15	0.28	2.0%	1.9%
Daytime	I	7.5	0.20	26.1%	40.7%
		15	0.34	22.1%	7.8%
	II	7.5	0.20	20.1%	40.7%
		15	0.39	18.1%	19.8%
	III	7.5	0.28	2.4%	1.9%
		15	0.42	1.2%	0.8%
Accumulated C, Tw = 15 min	Real Leak	5.2	0.14 to 0.96	0.5%	0.2%
Accumulated C, Tw = 16 min	Real Leak	5.2		11.8%	0.3%

Figure 1. Correlation between observed and simulated leaks with a uniform (top) and factorized demand (bottom) distribution and a leakage size of 7.5 m³/h. The higher the correlation (darker), the better the result. See section ‘Model-based leak localization’ in the methods section for details.

Figure 2. Highest accumulated correlations in the uniform (top) and factorized model (bottom) from the real leak.

Figure 3. Percentage of false positive (FP) nodes (in black) versus District Metered Area (DMA) inflow (gray) on 8 December 2015 with uniform (graphs A and C) and factorized demand distribution (graphs B and D), time analysis step and time window 15 min (graphs A and B), the time analysis step 15 min and time window 60 min (graphs C and D).