Performance Analyses of Differential Evolution Algorithm Based on Dynamic Fitness Landscape

2. Related Work

Recently, fitness landscape is used in many aspects such as search-based software testing problems (Aleti, Moser & Grunske, 2016) producing approximate solutions of a combinatorial optimization problem through climbing combinatorial fitness landscapes (Basseur & Ffon, 2015). Different evolutionary algorithms are improved from different aspects based on traditional algorithms in order to attain better performance including faster convergence speed, fewer parameters, less running time and so on. But here, we are not going to focus on how to improve or propose new algorithms. On the contrary, we are going to employ fitness landscape analysis to study the performance of differential evolution algorithm in many ways. The concept of fitness landscape has been adopted widely in recent years in many fields. (Bolshakov, Pitzer, & Affenzeller, 2011)) considered applying fitness landscape analysis in simulation optimization for meta-optimization purposes and new insights are obtained in the field of fitness landscapes analysis for stochastic problems. (Malan & Engelbrecht, 2013b) proposed using ruggedness, funnels and gradients in the fitness landscapes analyzed and evaluated the performance and the effect of traditional particle swarm optimization (PSO). According to experimental results, these three metrics have valued as part-predictors of PSO performance on unknown problems if used in conjunction with measures approximating other features that have been linked to problem difficulty for PSOs. Meanwhile, they investigated whether a link can be found between problem characteristics and algorithm performance for PSOs. But there are not new insights to show algorithm performance clearly.