Dynamic function optimisation with hybridised extremal dynamics

Abstract

Dynamic function optimisation is an important research area because many real-world problems are inherently dynamic in nature. Over the years, a wide variety of algorithms have been proposed to solve dynamic optimisation problems, and many of these algorithms have used the Moving Peaks (MP) benchmark to test their own capabilities against other approaches. This paper presents a detailed account of our hybridised Extremal Optimisation (EO) approach that has achieved hitherto unsurpassed results on the three standardised scenarios of the MP problem. Several different components are used in the hybrid EO, and it has been shown that a large proportion of the quality of its outstanding performance is due to the local search component. In this paper, the behaviour of the local search algorithms used is analysed, and the roles of other components are discussed. In the concluding remarks, the generalisation ability of this method and its wider applicability are highlighted.

Dynamic Function Optimization: The Moving Peaks Benchmark

“…As can be seen from the table, Moser and Chiong's new MMEO [45] and EO + HJ [43] have achieved hitherto unsurpassed results on Scenario 2 of the MP problem. Among the PSO-based solutions with comparable results, Yang and Li's clustering PSO [57] obtained the best offline error of 1.06 followed by Blackwell and Branke's PSO with anti-convergence [4] on an offline error of 1.80 from solving Scenario 2.…”

“…The EO-based approaches [43][44][45] perform poorly on single-peak landscapes. According to the authors, this is mainly because their local search has long distances to cross (incurring many function evaluations while the best-known solution is still poor) using steps that have been calibrated for smaller "mountains".…”

“…In a concurrent attempt to improve the local search phase of MMEO, Moser and Chiong [45] presented a new MMEO algorithm by systematically exploring appropriate step length sequences and removing unnecessary redundant steps in its local search procedure. During the development of the original MMEO, it became clear that the step lengths used for the local search were crucial for its performance.…”

“…It is therefore useful to attempt a move in the opposite direction, but only when following the previous move's direction has failed. The new local search in [45] addressed these issues. Extensive numerical experiments on different configurations of the MP problem confirmed that the improvement is significant, in which the new MMEO outperforms all other available solutions in the literature.…”

An adaptive multi-population artificial bee colony algorithm for dynamic optimisation problems

“…-Dynamic function optimization with hybridized extremal dynamics (EO + HJ) [30] -A competitive clustering particle swarm optimizer for dynamic optimization problems (CCPSO) [31].…”