A comparison of swarm intelligence algorithms for structural engineering optimization: Fid-Bau Portal

A comparison of swarm intelligence algorithms for structural engineering optimization

Parpinelli, Rafael S. / Teodoro, Fabio R. / Lopes, Heitor S.

In this paper, we presented results verifying the differences in exploitation and exploration balance in different SI algorithms performing an unbiased comparison over three difficult structural engineering problems. The analysis of the statistical results indicated that no algorithm performed best for all the problems, except for the welded beam design problem. Notwithstanding, different results for each problem were achieved, suggesting that an algorithm can be the best for a problem but not for other one. This happens mainly due the different exploration/exploitation strategies of the algorithms, which direct the search in different ways. In addition, each problem has different fitness landscapes (represented by its cost function and its constraints), which, by itself, may lead the same algorithm to perform differently for each problem. Such results are in accordance with the no free lunch theorem, which states that it is not possible to point out which is the best optimization algorithm without considering a specific problem. Genetically speaking, the results showed that it is not possible to point which algorithm is more efficient for generic classes of problems, such as the constrained optimization problems approached here. For real-world problems, both the quality of solution and the computational cost are important issues. In this work, the Pareto plot was found to be a useful tool for comparing the performance of algorithms considering such a multicriteria approach. The explosion procedure used in PSO and ABC tend to increase significantly the number of function evaluations without, however, improving significantly the quality of solutions. This is contrary to what has been cited in the literature about this procedure. We speculate that the reason for this unexpected behavior is the parameters that regulate the frequency of explosion. Therefore, instead of using the standard values, a fine-tuning of such parameters can lead to better results. On the basis of the results presented, we consider some combinations between algorithms as an alternative to solve these and other problems even more efficiently. Recent literature has indicated that the use of hybrid evolutionary systems working in a cooperative way can perform better than the use of single algorithms. A possible approach for this is to form a pipeline, passing the results from one algorithm to another or in parallel, choosing the best result from the parallel runs of several algorithms, or by having communication between the algorithms and each other. Therefore, we believe that hybrid/cooperative optimization strategies for highly constrained engineering problems are a promising future research.