Optimal Reliability-Based Design of Bulk Water Supply Systems

Abstract

DJ :;sc rt(l 1I UIl s uhnlltt ed In p a rt i~l fulfi lm en t o~' th e reqlllrcrnen ts fu r tht., degree ofM~s t e r ofSClence In Enginee ring AbstractBulk water supply systems are usually designed according to deterministic design guidelines. In South Africa, design guidelines specify that a bulk storage reservoir should have a storage capacity of 48 hours of annual average daily demand (AADD), and the feeder pipe a capacity of 1.5 times AADD (CSIR, 2000). Nel & Haarhoff (1996) proposed a stochastic analysis method that allowed the reliability of a reservoir to be estimated based on a Monte Carlo analysis of consumer demand, fire water demand and pipe failures. Van Zyl et al. (2008) developed this method further and proposed a design criterion of one failure in ten years under seasonal peak conditions.In this study, a method for the optimal design of bulk water supply systems is proposed with the design variables being the configuration of the feeder pipe system, the feeder pipe diameters (i.e. capacity), and the size of the bulk storage reservoir. The stochastic analysis method is applied to determine a trade-off curve between system cost and reliability, from which the designer can select a suitable solution.Optimisation of the bulk system was performed using the multi-objective genetic algorithm, NSGA-II. As Monte Carlo sampling can be computationally expensive, especially when large numbers of simulations are required in an optimisation exercise, a compression heuristic was implemented and refined to reduce the computational effort required of the stochastic simulation. Use of the compression heuristic instead of full Monte Carlo simulation in the reliability analysis achieved computational time savings of around 75% for the optimisation of a typical system.Application of the optimisation model showed that it was able to successfully produce a set of Pareto-optimal solutions ranging from low reliability, low cost solutions to high reliability, high cost solutions. The proposed method was first applied to a typical system, resulting in an optimal reservoir size of approximately 22 h AADD and feeder pipe capacity of 2 times AADD. This solution achieved 9% savings in total system cost compared to the South African design guidelines. In addition, the optimal solution proved to have better reliability that one designed according to South African guidelines.A sensitivity analysis demonstrated the effects of changing various system and stochastic parameters from typical to low and high values. The sensitivity results revealed that the length of the feeder pipe system has the greatest impact on both the cost and reliability of the bulk system. It was also found that a single feeder pipe is optimal in most cases, and that parallel feeder pipes are only optimal for short feeder pipe lengths.The optimisation model is capable of narrowing down the search region to a handful of possible design solutions, and can thus be used by the engineer as a tool to assist with the design of the final system.

Speeding up stochastic analysis of bulk water supply systems using a compression heuristic

“…Optimisation of the bulk system could take up to 269 h on a 2.66 GHz Intel Core 2 Quad processor using the full stochastic analysis method as presented by Van Zyl et al (2008), whereas the compression heuristic reduces the simulation time by 75% to approximately 66 h. This allowed stochastic analysis to be successfully incorporated in a genetic algorithm optimisation study (Chang and Van Zyl, 2012) …”

“…In a recent study (Chang and Van Zyl, 2012), stochastic analysis was used in combination with genetic algorithms to optimise the design of bulk water supply systems. The optimisation variables were the storage tank capacity, feeder pipe configuration and feeder pipe diameter.…”

Optimizing Conduit Hydropower Potential by Determining Pareto-Optimal Trade-Off Curve

“…According to Chang and Van Zyl [90], various authors have analyzed the reliability of bulk water supply systems using frequency durations, Markov chains, and Monte Carlo approaches.…”