An Informed Convergence Accelerator for Evolutionary Multiobjective Optimiser

Abstract

A novel optimisation accelerator deploying neural network predictions and 
objective space direct manipulation strategies is presented. The concept of 
directing the search through the use of 'mirage' solutions is introduced and 
investigated. The accelerator is meant to be a portable component that can 
be plugged into any stochastic optimisation algorithm, such as genetic algorithms. 
The purpose of the new component termed as the Informed Convergence Accelerator 
(ICA) is to enhance the search capability, convergence extent and most especially the 
speed of convergence of the hosting stochastic global optimisation technique. ICA was 
hybridized with the Non-Dominated Sorting Genetic Algorithm (NSGA-II). Enhanced 
results were achieved demonstrating the utility of the introduced component.