Solving microelectronic thermal management problems using a generalized spiral optimization algorithm

摘要

Metaheuristics have risen as an approach for addressing diverse optimization problems by mimicking biological processes. They have proven to be effective in different fields and problems, which has skyrocketed their popularity. A recent proposal, the spiral optimization algorithm (SOA), is based on the logarithmic spiral behavior that appears in several natural scenarios. Variants of this deterministic SOA (DSOA) have emerged, seeking to improve its performance. One of them is the stochastic SOA (SSOA), which transforms a deterministic path into a random path. In this work, we make two contributions. First, we present a generalized version of the algorithm that includes the DSOA and SSOA. We use it to study the effect of allowing a random ‘reflection’ in the rotation angle of the spirals. In our proposed approach, a ‘reflection’ entails replacing the rotation angle (?) with its supplementary angle (180∘− ?) in the current iteration. Thus, the ‘reflection’ allows for increased diversity when exploring the search domain. To test this idea, we use several test functions, including the CEC2005 benchmark, in multiple dimensions. Finally, we use this reflection-based optimization of the stochastic spiral algorithm (ROSSA) to solve a microelectronic thermal management problem from the literature and compare its performance against some recently reported values. The data reveal that adding the proposed coefficient leads to a statistically significant improvement in performance. Most ROSSA configurations outperform all tested settings of the DSOA and SSOA, for virtually all problems. Hence, we firmly believe that the proposed generalization is adequate and that random ‘reflections’ help improve the performance of the DSOA, without increasing its computational burden.