ABSTRACT
ABSTRACT: Performance of water distribution systems (WDS) deteriorates with time. As a consequence, these systems should be rehabilitated periodically. Current state-of-the-art methodologies tend to solve the problem of least cost WDS design/rehabilitation using evolutionary optimisation methods, genetic algorithms (GAs) in particular. However, in all these approaches, a GA is linked to a deterministic WDS simulation model neglecting a fundamental fact that a number of uncertainties exist in the decision making process. To reflect this, a stochastic least cost design problem is formulated and solved here. The objective is to minimise total design costs subject to a pre-specified level of design reliability. The optimisation problem is solved by linking a GA to a stochastic WDS hydraulic model. Each uncertain nodal demand modelled is assigned a probability density function (PDF) in the problem formulation phase. The corresponding PDFs of the computed nodal heads are calculated using the Latin Hypercube sampling technique. However, rather than using a large number of samples for each fitness evaluation throughout the search process, an alternative approach with a significantly reduced level of sampling is developed. The robust optimisation methodology is tested on a case study. The results obtained indicate that the new methodology is capable of identifying robust (near) optimal solutions despite significantly reduced computational effort compared to fitness evaluation using full sampling.
