A Novel Crossover-First Differential Evolution Algorithm with Explicitly Tunable Mutation Rates for Evolutionary-Based Global Optimization
Jason Teo, Kim-On Chin, Shaliza Hayati A. Wahab, Azali Saudi, Siti Hasnah Tanalol · 2017
Differential Evolution (DE) is currently one of the most popular evolutionary-based global optimization algorithms being simple to understand and implement as well as having fast convergence and robustness across a wide range of problems. Although it is classed as an evolutionary algorithm (EA), its genetic operations are atypical of such classes of algorithms. EAs typically perform crossover followed by mutation where both operations have an explicitly tunable rate of operation. However in DE, the mutation operation is conducted before the crossover operation. Moreover, although DE has a crossover rate, it does not have a mutation rate; rather it mandatorily mutates every gene in its chromosome essentially performing a 100% rate of mutation. Following this line of observation, we proceeded to experiment with a novel version of DE where the crossover and mutation operations are reversed to mimic typical EAs as well as to add in an explicitly tunable mutation rate. We have found that this simple and intuitive yet previously unexplored modification to DE is able to improve its performance, particularly in more complex search spaces with highly non-uniform fitness landscapes. Non-parametric tests show that the improvements are statistically significant