Parameter Tuning Strategies for Metaheuristic Methods Applied to Discrete Optimization of Structural Design

  • 2021-10-12 17:34:39
  • Iván Negrin, Dirk Roose, Ernesto Chagoyén
  • 0

Abstract

This paper presents several strategies to tune the parameters ofmetaheuristic methods for (discrete) design optimization of reinforced concrete(RC) structures. A novel utility metric is proposed, based on the area underthe average performance curve. The process of modelling, analysis and design ofrealistic RC structures leads to objective functions for which the evaluationis computationally very expensive. To avoid costly simulations, two types ofsurrogate models are used. The first one consists of the creation of a databasecontaining all possible solutions. The second one uses benchmark functions tocreate a discrete sub-space of them, simulating the main features of realisticproblems. Parameter tuning of four metaheuristics is performed based on twostrategies. The main difference between them is the parameter controlestablished to perform partial assessments. The simplest strategy is suitableto tune good `generalist' methods, i.e., methods with good performanceregardless the parameter configuration. The other one is more expensive, but iswell suited to assess any method. Tuning results prove that Biogeography-BasedOptimization, a relatively new evolutionary algorithm, outperforms othermethods such as GA or PSO for such optimization problems, due to its particularapproach of applying recombination and mutation operators.

 

Quick Read (beta)

loading the full paper ...