When conducting a preliminary search across an engineering design space using an evolutionary search method such as the Genetic Algorithm (GA) it is important to achieve the correct balance between exploration and exploitation. If search is too exploratitive, progress may rapidly degenerate into a random walk where the benefits of evolutionary earch are quickly lost. Conversely, if the degree of exploitation is too high, premature convergence may result, with significant areas of the search space remaining largely under explored. This paper introduces a number of COGA strategies developed to better explore and exploit the search space thereby promoting its subsequent decomposition into regions of high performance. Each technique is compared with Variable Mutation COGA (vmCOGA) upon a multi-dimensional high modality test function.
展开▼
