Stochastic Water Demand Modelling PDF Download

Typical network modeling in distribution system analysis assumes the water demands are known and constant over 1 hour. However, as utilities move towards "all-pipe" network models, continuing to ignore the stochastic water demands may not adequately account for the impacts of demand variability on the underlying transport and water quality characteristics. The objective of this research is to evaluate the potential impacts of different levels of temporal aggregation of water demands on the underlying hydraulic, transport, and water quality simulations for both a small network and a large "all-pipe" network system. A non-homogeneous Poisson Rectangular Pulse model was used to generate stochastic water demands aggregated at 1-min, 10-min, and 1-hr time steps, and linked with EPANET to perform hydraulic and water quality simulations. The impacts of the three temporal aggregations of water demands were evaluated with respect to: 1) hydraulics by evaluating pressure and flow rate variability; 2) transport and water quality characteristics using "conservative chemical intrusion" events and evaluating transport times and cumulative mass. Additional studies were performed to interpret the chemical analysis within a risk analysis framework, and investigated the impact of temporal aggregation with different injection durations. For a small skeletonized network, results showed for main trunk lines, the demand variability had little influence on the flow rate, chemical concentration, and risk assessment. However, as dead end nodes or pipes were analyzed, there was an increase in the flow rate variability with decreasing temporal aggregation that impacted the chemical concentration time series and risk assessment by altering the travel paths and times. For the large "all-pipe" network model, the results illustrated a greater frequency in flow reversals as well as meaningful differences in the initial arrival time and half-mass arrival time of chemical with decreasing temporal aggregation scales. Upon analyzing the Monte Carlo ensemble of results, the majority of nodes that resulted in meaningful initial chemical arrival time differences tended to be located at the edges of the network. With respect to the half-mass arrival time, a spatial analysis indicated there were a number of nodes within a blending region that had meaningful differences within a majority of the Monte Carlo realizations. Finally, a comparison of the stochastic water demand simulations to the original deterministic simulation indicated that the intra-hour variability resulting from the stochastic demands was more important than the variability resulting from the temporal aggregation scale. The results associated with a risk assessment also showed meaningful differences in the time until 1% and 50% of the population at a node infected by a toxic chemical between the shorter temporal aggregation scales and the 1-hr case. The impacts of injection duration illustrated the injection duration had little impact on the initial chemical arrival time. However, decreasing the injection duration had greater impacts on the differences of half-mass chemical arrival time and cumulative mass between the 10-min and 1-hr aggregation cases. These results indicate that there are portions of a distribution system where the typical network modeling assumptions may not be appropriate to adequately represent localized transport.