Venue: Proceedings of the 2020 Genetic and Evolutionary Computation Conference
Year: 2020
Paper: https://dl.acm.org/doi/10.1145/3377930.3390204
Abstract
The application of Evolutionary Algorithms (EAs) to real-world problems comes with inherent challenges, primarily the difficulty in defining the large number of considerations needed when designing complex systems such as Water Distribution Networks (WDN). One solution is to use an Interactive Evolutionary Algorithm (IEA), which integrates a human expert into the optimisation process and helps guide it to solutions more suited to real-world application. The involvement of an expert provides the algorithm with valuable domain knowledge; however, it is an intensive task requiring extensive interaction, leading to user fatigue and reduced effectiveness. To address this, the authors have developed methods for capturing human expertise from user interactions utilising machine learning to produce Human-Derived Heuristics (HDH) which are integrated into an EA’s mutation operator. This work focuses on the development of an adaptive method for applying multiple HDHs throughout an EA’s search. The new adaptive approach is shown to outperform both singular HDH approaches and traditional EAs on a range of large scale WDN design problems. This work paves the way for the development of a new type of IEA that has the capability of learning from human experts whilst minimising user fatigue.