A novel evolutionary algorithm for global optimization for which derivative information is not required, that rather deals with function value information only, is addressed in this paper. The focus of this method is a dynamic partition search algorithm (DPSA). The authors base their approach on population, like methods using stochastic searches in nature, in which there are two components of the search: initialization and evolution processes. The initialization process involves random methods, which generate an initial population, and develop a “leader,” which provides the best result thus far. The evolutionary group is based upon Cauchy and Gaussian distributions that search the region around the “leader.” The paper provides details of the results of 24 benchmark functions of up to 10,000 dimensions. A variety of alternative methods are compared with the DPSA approach, such as cooperative coevolving particle swarm optimization (CCPSO2) and other existing differential evolution (DE) algorithms. Detailed results of these benchmarks and results demonstrate good scalability on a range of problems in many cases.
The paper begins with an introduction that surveys a range of evolutionary methods (DEs) developed over past decades. A section follows this that explains a DPSA in detail. The third section, “Experimental Studies,” provides detailed descriptions of these studies, and is followed by “Results and Analysis” section. This section includes detailed explanations, as well as tabular and graphical results. A final section concludes the paper. This is an excellent paper that provides considerable insight to the performance and results of the method.