Calibration of water distribution network of the Ramnagar zone in Nagpur City using online pressure and flow data

A water distribution network (WDN) is a network of interconnected pipes involving several components. Each network is unique in source, layout, topography of service area, pipe material, valves and meters and consumer connections. Hydraulic simulation model are widely used for predicting the behavior of the network in terms of nodal heads and pipe discharges. However, disagreement between model-simulated behavior and actual field behavior often exists because of uncertainty in input parameters used in the model. Before any intended use of the WDN model, it must be ensured that the model would predict, with reasonable accuracy, the behavior of the network in real time. The process of adjusting the value of uncertain parameters in the network modeling to match values of modeled parameters with those observed in the field is called calibration of the model (Walski 1983a; Ormsbee and Wood 1986; Bhave 1988). Walski (1983b) precisely defined calibration as “a two step process consisting of: (1) comparison of pressures and flows predicted with observed pressures and flows for known operating conditions (i.e., pump operation, tank levels, pressure reducing valve (PRV) settings); and (2) adjustment of the input data for the model to improve agreement between observed and predicted values.”

Calibration is a challenging task, as a large numbers of input parameters are involved (e.g., nodal elevations and demands including water losses; pipe length, diameter and roughness coefficients; the rate of water supply and water level at the source; pump characteristics; valve settings, etc.) and depends on the accuracy of these parameters gathering and preparation. Therefore, it is desirable that some of these parameters are measured accurately during field observations so that there are few uncertain parameters requiring adjustments. Further, instead of developing different flow conditions to get the flow and pressure readings, a computerized flow and pressure measurement with a data logger provides fairly accurate flow and pressure at various operating conditions in a day. Field observations and measurements provide reliable data regarding pipe length and material, ground elevations, operational settings and status of valve and water supply and level at the source node. The nodal demands are obtained from the consumer usage information, i.e., from billing data, and non-revenue water (NRW) is adjusted. The corrosion and deposition processes over a time after the pipe has been installed make difficult to determine actual pipe diameter. Therefore, normal pipe diameter obtained during field observation is used in the hydraulic model and the roughness coefficient is adjusted to compensate for the change in the diameter. In general, therefore data regarding the pipe length, diameter and material, nodal elevation and demands adjusting NRW and water level and supplies at the source node obtained during field observations are considered fairly reliable and do not need any adjustment during calibration. However, the pipe roughness coefficients are less reliable and therefore may need adjustment during calibration.

Several methods have been suggested to simultaneously adjust pipe roughness coefficients and nodal demands (Bhave and Gupta 2006), or pipe roughness coefficient only with the assumption that nodal demands are fairly accurate and do not need any adjustment. The calibration of the model could be carried out through explicit approaches (e.g., Ormsbee and Wood 1986; Bhave 1988; Boulos and Wood 1990; Todini 1999) in which unknown pipe roughness coefficients are obtained by framing hydraulic equations using the observed flow and pressure values. The same number of equations are required as the number of unknowns. The other types of methods for calibration of the model are implicit approaches (e.g., Ormsbee 1989; Lansey and Basnet 1991; Datta and Sridharan 1994; Greco and Giudice 1999; Greco and Di Cristo 1999; Lansey et al. 2001; Kapelan 2002; Lingireddy and Ormsbee 2002; Bascià and Tucciarelli 2003; Kapelan et al. 2007; Koppel and Vassiljev 2009; Alvisi and Franchini 2010), in which field-observed and measured parameters are treated as known parameters and directly used in the model analysis. The implicit approach requires that the number of flow and pressure measurements exceed the number of unknowns. The implicit problem formulation can be solved using optimization techniques [e.g., nonlinear programming (NLP) or evolutionary technique like GA].

Reasonable agreement of the model is usually judged in terms of the differences in the observed and predicted pressures or heads at the test nodes (e.g., Cesario and Davis 1984; Eggener and Polkowski 1976; Rahal et al. 1980; Walski 1983b), and depends on the accuracy of the data set and the effort and cost the model user is prepared to spend in fine-tuning the model. Walski (1983b) states that an average difference of ± 1.5 m with a maximum value of ± 5.0 m for a good data set and the corresponding values of ± 3.0 m and ± 10 m for a poor data set would be a reasonable target. Cesario and Davis (1984) state that the model can be calibrated to an accuracy of 3.5–7 m.

Instead of judging the calibration through only differences and/or ratios of the observed to the predicted pressure or head loss differences between the test nodes and nearby boundary locations (reservoirs/tanks) having known pressures or heads (Walski 1986), it is preferable to judge the calibration through minimizing the summation of the sum of square of difference between the measured and simulated value of pressure heads at test nodes as done in most of the implicit approaches The Research Committee on Water Distribution Systems of the American Water Works Association (1974) states that “the major source of error in simulation of contemporary performance will be in the assumed loadings distribution and their variations”. On the other hand, Eggener and Polkowski (1976) state that the weakest piece of input information is not the assumed loading condition, but the pipe friction factor. From these conflicting views, it is clear that, in some cases, the adjustment in the pipe head-loss coefficients may be more predominant than the adjustment in the nodal consumptions, while the opposite may be the situation in other cases (Shamir and Howard 1977). Walski (1986) suggested consideration of multiple demand loadings for reliable model calibration, which can be generated through pump on/off conditions or excessive withdrawal at one or few nodes by opening fire hydrants. The novelty of the study herein is the use of online data of flow and pressures that provided a set of observations under various loadings. This paper demonstrates that with the online pressure and flow data at few observational points in the network, it is possible to accurately calibrate a model of a real WDN by adjusting the pipe roughness coefficients only. This is achieved by providing accurate inputs about pipe lengths, diameters and materials, nodal elevations, nodal demands, valve status and its opening. The importance of the valve is recognized, as these are used for isolation as well as pressure and flow control purpose by throttling. The usual way of considering head loss through them as minor loss and clubbing their effects in adjusting pipe roughness coefficient results in a very absurd value of pipe roughness coefficients.