Impact of Starting Position and Searching Mechanism on the Evolutionary Algorithm Convergence Rate

摘要

Evolutionary algorithms (EAs) have been used extensively to find globally optimal solutions for water distribution system (WDS) optimization problems. However, as these algorithms are being applied to increasingly complex systems, computational efficiency is becoming an issue, and hence approaches that enable near-optimal solutions to be identified within reasonable computational budgets have received increasing attention. One of these approaches is the initialization of EAs in a manner that accounts for domain knowledge of WDS design problems. Although the effectiveness of these initialization approaches has been studied previously, the impact of algorithm searching behavior on the speed with which near-optimal solutions can be found has not yet been examined. To this end, this study aims to investigate the relative impact of different algorithm initialization methods and searching mechanisms on the speed with which near-optimal solutions can be identified for large WDS optimization problems. Fitness function and run-time behavioral statistics are used for this purpose. The results show that both the starting population and algorithm searching mechanism have an impact on the speed with which near-optimal solutions are identified. The fitness function and run-time behavioral statistics indicate that EA parameterizations that favor exploitation over exploration enable near-optimal solutions to be identified earlier in the search, which is due to the big bowl shape of the fitness function for all of the WDS problems considered. Using initial populations that are informed by domain knowledge further increases the speed with which near-optimal solutions can be identified.