Cooperation Coevolution Differential Evolution with Gradient Descent Strategy for Large Scale

Abstract

In order to better solve the large scale optimization problem, we propose a cooperation coevolution differential evolutionary (CCDE) algorithm with a gradient descent strategy (GDS). The GDS based CCDE algorithm (CCDE/GDS) benefits the solution of large scale optimization problems in two critical aspects. Firstly, the optimization turned out to be far less time consuming due to that GDS is helpful for guiding the search direction on the globally best individual position. More importantly, the GDS is controlled by an elastic operator to be carried out only when the globally best individual has been trapped, making the algorithm fast respond to the large scale evolutionary environment. Secondly, GDS was reported in the literature to approximate the local best value on most object functions. Therefore, the GDS used in CCDE can promote the globally best individual position to more promising region when it is trapped into local optimum, so as to achieve high accuracy. We designed experiments on CEC2010 benchmark functions for evaluating our newly proposed algorithm, which shows that the proposed algorithm and modified framework can obtain very competitive results on the large scale optimization problem efficiently.