DYN_CBC: Dynamic Adjustment of Context Based Clearing for Advanced GA Niching

Abstract

The inability of GA to find multiple global maxima of a multimodal function has been a basic limitation. Many niching and diversity preserving techniques have been suggested to overcome it. Clearing is one of the common approaches proposed. Due to its promising performance it has attracted researchers to further enhance it. CBCI (Context Based Clearing) is an updated version of standard clearing procedure. This approach made use of context information to prevent clearing of some candidates that may lead to undetected optima. In this paper three advanced dynamic versions of CBC are proposed (CBCII, CBCIII, and CBCIV) to further enhance the clearing process. They are modified versions of CBCI, each by a different approach. The proposed techniques were tested using the M7 function; a massively multimodal deceptive optimization function typically used for testing the efficiency of finding global optima in a search space. The results are compared with those of CBCI. Results show that the new versions discover more optima in earlier generations than CBCI while having a comparable complexity with CBCI. Compared to other reported clearing techniques such as Spatially-Structured EAs (SSEAs) using ring topology, LC (Local clearing) and direct crowding (CD) CBCIV finds optima much faster. the suggested techniques outperform them as regards complexity and manages to find global M7 optima much faster