Comparison of Six Parameter-Less Evolutionary Algorithms for the Optimal Design of Water Distribution Networks

Water distribution networks (WDNs) play a crucial role in the development of any society. These are expensive infrastructures, and therefore cost optimization is desirable. In the previous decade, the use of Evolutionary Algorithms (EAs) as an optimization tool has grown substantially and several new algorithms have been proposed. EAs work on metaheuristics based on some natural phenomenon, which may be physical, chemical, or biological inspired, or may be based on some other strategy. These algorithms require the use of some parameters to mimic the behavior of adopted metaheuristic. These parameters are problem-specific and require fine-tuning either by the user each time a new problem is solved or done internally on its own during execution of the algorithm to avoid the necessity of fine-tuning by the user. Obviously, internal fine-tuning reduces computational efforts. Thus, EAs can be broadly classified as parameter-based and parameter-less algorithms depending on the need for fine-tuning of parameters by users. In this chapter six parameter-less EAs namely Jaya, Rao-I, Rao-II, Rao-III, Black hole optimization (BHO), and Teaching learning-based optimization (TLBO) are described and compared for the optimal pipe sizing problem. Among the six EAs, BHO and TLBO were not applied earlier for the optimization of WDN. All six EAs are applied on a well-known benchmark network taken from the literature to compare the optimal cost and the required number of functional evaluations (hydraulic analysis) to reach the best cost solution by fixing the population size and termination criteria (a fixed set of number of iterations). Among the six EAs, it was found that Rao-II outperformed the other five EAs. BHO is observed to fail to converge to the optimal results for the example WDN.