Benchmark Generator for the IEEE WCCI-2014 Competition on Evolutionary Computation for Dynamic Optimization Problems Dynamic rotation peak benchmark generator (DRPBG) and Dynamic composition benchmark generator (DCBG)

Many realworld optimization problems are dynamic optimization problems (DOPs), where changes may occur over time regarding the objective function, decision variable, and constraints, etc. DOPs raise big challenges to traditional optimization methods as well as evolutionary algorithms (EAs). The last decade has witnessed increasing research efforts on handling dynamic optimization problems using EAs and other metaheuristics, and a variety of methods have been reported across a broad range of application backgrounds. In order to study the performance of EAs in dynamic environments, one important task is to develop proper dynamic benchmark problems. Over the years, researchers have applied a number of dynamic test problems to compare the performance of EAs in dynamic environments, e.g., the “moving peaks” benchmark (MPB) proposed by Branke [1], the DF1 generator introduced by Morrison and De Jong [7], the singleand multi-objective dynamic test problem generator by dynamically combining different objective functions of exiting stationary multi-objective benchmark problems suggested by Jin and Sendhoff [2], Yang and Yao’s exclusive-or (XOR) operator [11, 12, 13], Kang’s dynamic traveling salesman problem (DTSP) [3] and dynamic multi knapsack problem (DKP), etc. Although a number of DOP generators exist in the literature, there is no unified approach of constructing dynamic problems across the binary space, real space and combinatorial space so far. This report uses the generalized dynamic benchmark generator (GDBG) proposed in [4], which construct dynamic environments for all the three solution spaces. Especially, in the real space, we introduce a rotation method instead of shifting the positions of peaks as in the MPB and DF1 generators. The rotation method can overcome the problem of unequal challenge per change for algorithms of the MPB generator, which happens when the peak positions bounce back from the boundary of the landscape. Based on our previous benchmark generator for the IEEE CEC’12 Competition on Dynamic Optimization [5], this report updates the two benchmark instances where two new features have