Augmented evolutionary intelligence

Abstract

Evolutionary Algorithms (EAs) have been employed for the optimisation of both theoretical and real-world problems for decades. These methods although capable of producing nearoptimal solutions, often fail to meet real-world application requirements due to considerations which are hard to define in an objective function. One solution is to employ an Interactive Evolutionary Algorithm (IEA), involving an expert human practitioner in the optimisation process to help guide the algorithm to a solution more suited to real-world implementation. This approach requires the practitioner to make thousands of decisions during an optimisation, potentially leading to user fatigue and diminishing the algorithm's search ability. This work proposes a method for capturing engineering expertise through machine learning techniques and integrating the resultant heuristic into an EA through its mutation operator. The human-derived heuristic based mutation is assessed on a range of water distribution network design problems from the literature and shown to often outperform traditional EA approaches. These developments open up the potential for more effective interaction between human expert and evolutionary techniques and with potential application to a much larger and diverse set of problems beyond the field of water systems engineering.