Modeling activities, specifically model calibration, need sensor measurements. Since data collection is expensive and time-consuming, ineffective measurements should be avoided. For this reason, it is important to have a method for designing the field data collection that will give the most useful information. While sampling design problems have been studied more in ground- water systems, there are only few applications to water distribution networks. In this paper three different sensitivity-based methods for selecting the worthwhile sensor location in water distribution network are compared. The covariance matrix method, previously applied to identify the most useful loading condition under which to take measurements, is adapted to a sampling design problem. Since this method considers the correlation between the coefficients of the sensitivity matrix of the state variables with respect to calibration parameters, it has a higher computational cost. The three methods are applied to a hypothetical roughness calibration problem. The results show that there are no marked differences between the three methods, while the difference between poor and good sampling design is significant when the number of sensors is limited. This particular result remarks the importance of using a procedure for selecting sensor location
Sampling design for water distribution networks
DI CRISTO, Cristiana
2003-01-01
Abstract
Modeling activities, specifically model calibration, need sensor measurements. Since data collection is expensive and time-consuming, ineffective measurements should be avoided. For this reason, it is important to have a method for designing the field data collection that will give the most useful information. While sampling design problems have been studied more in ground- water systems, there are only few applications to water distribution networks. In this paper three different sensitivity-based methods for selecting the worthwhile sensor location in water distribution network are compared. The covariance matrix method, previously applied to identify the most useful loading condition under which to take measurements, is adapted to a sampling design problem. Since this method considers the correlation between the coefficients of the sensitivity matrix of the state variables with respect to calibration parameters, it has a higher computational cost. The three methods are applied to a hypothetical roughness calibration problem. The results show that there are no marked differences between the three methods, while the difference between poor and good sampling design is significant when the number of sensors is limited. This particular result remarks the importance of using a procedure for selecting sensor locationI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.