Introducing intervention targeting into estimation of distribution algorithms
Geoffrey Neumann, David R. Cairns · 2012
This paper introduces a new hybrid Genetic Algorithm (GA) crossover approach, Targeted EDA (TEDA), that combines a targeted intervention principle with Estimation of Distribution Algorithms (EDA) to solve optimal control problems. The approach is suited to tasks where the number of interventions used is an important part of solution fitness and includes problems such as cancer chemotherapy scheduling. Fitness Directed Crossover (FDC) is a modified GA crossover method that actively drives the number of selected control interventions towards those of a fitter individual. EDA are able to find fit solutions by discovering patterns within a population of selected individuals. TEDA uses FDC to select a suitable number of interventions to use while using an EDA based approach to select which interventions to set. Results suggest that by combining the two approaches, TEDA is able to outperform both EDA and FDC on a sample optimal control problem.