This paper proposes the application of three different methods for preserving the correlation between duration and intensity of synthetically generated water demand pulses. The first two methods, i.e., the Iman and Canover (1982) method and the Gaussian copula (Nelsen, 1999) respectively, are derived from the scientific literature of statistics, though they had never been applied to the context of demand pulse generation. The third is novel and comes from a variation in the second. Poisson models fitted with the methods are applied to the case study of one of the Milford households (Buchberger et al., 2003), with parameters being obtained with the method of the moments. Comparisons are made with another method previously proposed in the scientific literature (Creaco et al., 2014), which has the disadvantage of being applicable only when the two variables are represented by either the normal or the lognormal probability distribution. The applications show that the three methods studied in this paper yield results that are consistent to the measurements and are similar to those of the previous method in the case where this latter can be applied. The three methods prove to be effective also when the other cannot be applied (durations and intensities not represented either normally or log-normally). An insight is finally provided into the extent to which pulse generation models with or without correlated durations and intensities are affected by a change in the method for parameter estimation. In this context, the results indicate that the models with correlation are those which always provide better results, though the performance of models without correlation may be improved when other parameter estimation methods than the method of the moments are chosen.
Preserving duration-intensity correlation on synthetically generated water demand pulses
CREACO, Enrico Fortunato;ALVISI, Stefano;FRANCHINI, Marco;
2015
Abstract
This paper proposes the application of three different methods for preserving the correlation between duration and intensity of synthetically generated water demand pulses. The first two methods, i.e., the Iman and Canover (1982) method and the Gaussian copula (Nelsen, 1999) respectively, are derived from the scientific literature of statistics, though they had never been applied to the context of demand pulse generation. The third is novel and comes from a variation in the second. Poisson models fitted with the methods are applied to the case study of one of the Milford households (Buchberger et al., 2003), with parameters being obtained with the method of the moments. Comparisons are made with another method previously proposed in the scientific literature (Creaco et al., 2014), which has the disadvantage of being applicable only when the two variables are represented by either the normal or the lognormal probability distribution. The applications show that the three methods studied in this paper yield results that are consistent to the measurements and are similar to those of the previous method in the case where this latter can be applied. The three methods prove to be effective also when the other cannot be applied (durations and intensities not represented either normally or log-normally). An insight is finally provided into the extent to which pulse generation models with or without correlated durations and intensities are affected by a change in the method for parameter estimation. In this context, the results indicate that the models with correlation are those which always provide better results, though the performance of models without correlation may be improved when other parameter estimation methods than the method of the moments are chosen.I documenti in SFERA sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
