A Hybrid Stochastic Algorithm with Domain Reduction for Discrete Structural Optimization

Abstract

In recent years, many nature-inspired metaheuristic optimization algorithms have been proposed in an effort to develop efficient and robust algorithms. The drawback in most of them is the large number of simulations required to obtain good designs. To reduce the number of structural analyses to reach the best design, a new two-phase algorithm is proposed and evaluated. This hybrid algorithm is based on the well-known Harmony Search (HS) algorithm and recently developed Colliding Bodied Optimization (CBO). HS analyzes and improves one design in every iteration whereas CBO generates and analyzes a new population of designs in every iteration. Based on the observed behavior of these two algorithms, a Hybrid Harmony Search - Colliding Bodies Optimization (HHC) is proposed. The first phase of HHC uses the Improved Harmony Search (IHS) algorithm. A new design domain reduction technique is also incorporated in IHS that dramatically reduces the number of possible combinations of discrete variables. This improves the performance of the IHS algorithm. The second phase uses the Enhanced Colliding Bodies Optimization (ECBO). ECBO receives final designs from the first phase to enhance them further. This makes the second phase need fewer iterations in comparison with the ECBO alone. The performance of the proposed algorithms is evaluated using some benchmark discrete structural optimization problems, although the method is applicable to continuous-variable problems as well. The results show HHC with design domain reduction to be quite effective, robust, and needs a smaller number of structural analyses to solve optimization problems in comparison with IHS, ECBO, and some other metaheuristic optimization algorithms. HHC with design domain reduction is shown to be quite robust in the sense that different runs for a problem obtain the same final design. In comparison with HIS and ECBO, HHCD reduces the number of structural analyses to find the best design to less than half. This is an important feature that leads to better confidence in the final solution from a single run of the algorithm for a problem.