Direct Memory Schemes for Population-Based Incremental Learning in Cyclically Changing Environments

The population-based incremental learning (PBIL) algorithm is a combination of evolutionary optimization and competitive learning. The integration of PBIL with associative memory schemes has been successfully applied to solve dynamic optimization problems (DOPs). The best sample together with its probability vector are stored and reused to generate the samples when an environmental change occurs. It is straight forward that these methods are suitable for dynamic environments that are guaranteed to reappear, known as cyclic DOPs. In this paper, direct memory schemes are integrated to the PBIL where only the sample is stored and reused directly to the current samples. Based on a series of cyclic dynamic test problems, experiments are conducted to compare PBILs with the two types of memory schemes. The experimental results show that one specific direct memory scheme, where memory-based immigrants are generated, always improves the performance of PBIL. Finally, the memory-based immigrant PBIL is compared with other peer algorithms and shows promising performance.