Thursday 03 Dec 2020: Learning from Experience: Hyperheuristic Optimisation for Operations Research Problems
Ed Keedwell - University of Exeter, Computer Science
Optimisation problems exist in many areas of business including logistics, scheduling, design, vehicle routing etc. These have typically been solved using metaheuristics (e.g. evolutionary algorithms, swarm intelligence) which search the space of possible solutions using a fixed set of operators to explore the search space and exploit known good solutions. Hyperheuristics are an alternative approach that operate at the level above metaheuristics and adapt themselves using either online, or offline (usually machine learning) techniques to create a bespoke optimiser for the search problem at hand. In this talk, I will describe the learning optimisation approach that is central to hyperheuristics and will cover my group's research on the development of online and offline learning of sequences of heuristics to solve problems in the operations research and water distribution network design spaces. The talk will conclude by demonstrating that sequence-based hyperheuristics are capable of generating high-quality solutions to these problems and also can reveal important information about the mapping between the problem domain, the maturity of the optimisation process and the search strategy employed.