An new efficient evolutionary approach for dynamic optimization problems
Yong Liang · 2009
To improve the efficiency of the currently known evolutionary algorithms for dynamic optimization problems, we have proposed a novel variable representation allows static evolutionary optimization approaches to be extended to efficiently explore global and better local optimal areas in dynamic fitness landscapes. It represents a single individual as three real-valued vectors (x,σ,r)∈ Rn× Rn× R2in the evolutionary search population. The first vector x corresponds to a point in the n-dimensional search space (an object variable vector), the second vector describes the search step of x, while the third vector r represents the dynamic fitness value and the dynamic tendency of the individual x in the dynamic environment. σ and r are the control variables (also called strategy variables), which allow self-adaptation. The object variable vector x is operated by different genetic strategies according to its corresponding σ and r. As a case study, we have integrated the new variable representation into Evolution Strategy (ES), yielding an Dynamic Optimization Evolution Strategy (DOES). DOES is experimentally tested with 5 benchmark dynamic problems. The results all demonstrate that DOES outperforms other ES on dynamic optimization problems.