Composite particle optimization with hyper-reflection scheme in dynamic environments

Dynamic optimization is a challenging problem to the classic particle swarm optimization algorithms, it requires the optimizer not only to find the global optimal solution under a specific fitness landscape but also to track the trajectory of changing optima. This paper investigates a composite particle swarm optimizer, which presents a novel version of interactions among particles, to address dynamic optimization problems. A new composite particle generation approach based on the “fittest-oriented” principle is proposed, it creates each composite particle by one fitter particle from the swarm and other two particles randomly generated in its neighborhoods. In order to integrate valuable information for searching the changed optima, we introduce a scatter factor into the velocity-anisotropic reflection (VAR) scheme and a “fitness-and-distance” based pioneer particle identification (PPI) method. In addition, the composite particles interact with other particles in the swarm using an integral movement strategy, which aims to enhance the diversity of the swarm. Based on the experimental results in static landscapes, a hyper-reflection mechanism is introduced to enhance the efficiency of the VAR operator. Experimental results on the effect of the introduced schemes and user-specified parameters on DF1 problem provides a guideline for setting the involved parameters. Experimental comparisons with other state-of-art PSO variants and evolutionary computation algorithms on DF1 functions together with a suite of DOPs generated from the generalized dynamic benchmark generator (GDBG), which were used for the 2009 Competition on Evolutionary Computation in Dynamic and Uncertain Environments (ECiDUE), are also provided.