Variable-size Memory Evolutionary Algorithm: Studies on the impact of different replacing strategies in the algorithm's performance and in the population's diversity when dealing with dynamic environments



Diversity and memory are two major aspects when dealing with dynamic environments. The algorithms' adaptability to changes is usually dependent on these two issues. In this paper we investigate some improvements to a memory-based evolutionary algorithm already studied with success in dynamic optimization problems. This algorithm uses a memory and a population both with variable sizes and a biological inspired recombination operator to control the population's diversity. We propose two new replacing strategies to incorporate in the algorithm and we perform a comparative study with previous approaches. These replacing strategies allow the memory to grow in a more controlled manner, storing relevant information from the different environments. The results show that the adaptability of the algorithm improves through the time, proving that the stored information becomes useful in future situations. Combined with the conjugation operator the proposed schemes powerfully improve the effectiveness of the algorithm. We also evidenced that high diversity levels doesn't always mean better performance of Evolutionary Algorithms in dynamic environments.


Evolutionary Algorithms, Dynamic Environments, Memory, Diversity, Replacing Strategies


Evolutionary Optimization

TechReport Number

TR 2007/001

Cited by

No citations found