Efficient parallel evolutionary optimization algorithm applied to a water distribution system

Water distribution systems are key components of public infrastructure and it is essential to design, manage and rehabilitate them economically without compromising the required performance and regulatory standards. Evolutionary optimization algorithms such as genetic algorithms have become popular in providing optimal and near optimal solutions to various optimisation problems on water distribution systems. However, one of the main challenges associated with genetic algorithms in the optimization of water distribution systems is that they are time consuming when applied to problems on real-world water distribution networks with large numbers of pipes and multiple operating conditions. For example, in the optimisation of large water distribution systems, a single optimisation run may involve millions of hydraulic and water quality simulations that may take many days on modern powerful computers such as workstations. One way to address this difficulty is by utilising parallel computing methods. The aim of the research was to investigate the potential improvement in computational efficiency achievable through parallel computing by optimizing a design based on a real-world water distribution network with a large number of decision variables, large solution space and complex response surface. The results showed that the parallel algorithm developed was practical and found optimal and near-optimal solutions reliably and efficiently, and the solutions achieved were consistently competitive. The average speedup achieved on an eight-core workstation was 15, based on an optimization problem with hundreds of decision variables for a real-world water distribution network with extended period simulation.