EMPIRICAL ANALYSIS OF MEMETIC ALGORITHMS FOR CONCEPTUAL DESIGN OF. STEEL STRUCTURAL SYSTEMS IN TALL BUILDINGS
Department of Civil, Environmental and Infrastructure Engineering,
George Mason University,
Fairfax, VA, USA
E-mail: rkicinge@gmu. edu
This paper discusses the results of extensive design experiments in which memetic algorithms were applied to optimize topologies of steel structural systems in tall buildings. In these experiments, evolutionary algorithms were employed to determine optimal configurations of structural members (topology optimization) while the optimal cross-sections of members (sizing optimization) were found using continuous/discrete optimization algorithm implemented in SODA. The impact of all major evolutionary computation parameters on the performance of memetic algorithms was investigated. Two classes of complex structural design problems were considered: design of a wind bracing system in a tall building and design of the entire steel structural system in a tall building. The total weight of the structural system was assumed as the optimality criterion with respect to which the designs were optimized while satisfying all design requirements specified by appropriate design codes.
In the conducted experiments various key EC parameters and their values were considered having the largest impact on the performance of memetic algorithms. It was discovered that the type of EA, the rate of mutation operator, and the size of parent population were critical for the success of structural optimization processes. Specifically, evolution strategies produced on average significantly better results than genetic algorithms for the design problems considered in the paper. Also, low mutation rates, i. e. 0.025, resulted in best performance of memetic algorithms. Furthermore, small parent population sizes were generally preferred to large populations. For the simpler problem of conceptual design of a wind bracing system, optimal results were produced even when the population with a single member was used. In the case of the second and more complex design problem slightly larger population sizes were required consisting of 5 members.
Results of a large number of design experiments allowed formulating initial recommendations regarding optimal parameter settings for memetic algorithms for structural design applications. The experiments also produced a body of structural design knowledge, both quantitative and qualitative in nature. They identified regions of the design spaces in which high-performance solutions can be found. They also defined the ranges of the total weight of structural systems associated with high – performance solutions for both classes of design problems. Furthermore, significant qualitative differences between high-performance solutions have been identified. The structural shaping patterns exhibited by high-performing designs ranged from crossed macrodiagonal patterns composed of X bracings to irregular patterns consisting of various types of bracings.
Keywords: Memetic algorithms, evolutionary computation, structural design, conceptual design, tall buildings
Evolutionary computation (EC) is used for solving many complex problems in science and engineering. It allows conducting robust optimization and at the same time has modest requirements on the formulation of the problem to be solved. For example, it does not require continuous variables, differentiable objective functions, etc. Thus, it can be applied to many structural design problems, particularly those with discrete or symbolic variables, or objective functions with nonlinear and stochastic components as is frequently the case in conceptual design.
Parallel search conducted by a population (superset) of solutions is one of the key characteristics of evolutionary algorithms (EAs). It facilitates global exploration of the search space and helps EAs
M. Pandey et al. (eds), Advances in Engineering Structures, Mechanics & Construction, 277-288.
© 2006 Springer. Printed in the Netherlands.
escape local optima. In design spaces, these local optima frequently correspond to already known design concepts. Parallel search, however, also has an adverse impact on the ability of EAs to efficiently refine near-optimal solutions. EAs are usually not as good as local search algorithms in converging to the optimal solution once the optimal region of the search space has been found.
Thus, several researchers proposed hybrid algorithms combining excellent global exploration characteristics of EAs and efficient refinement capabilities of local search algorithms (Hoeffler et al., 1973; Moscato, 1989). These hybrid algorithms are called memetic algorithms (MAs) but other names like hybrid EAs, or Lamarckian EAs, have also been used in the literature. Even though several applications of MAs to structural design have been reported, e. g., (Hoeffler et al., 1973; Quagliarella and Vicini, 1998; Sakamoto and Oda, 1993), none of them had such a broad empirical scope as the computational studies described in this paper.
This paper discusses the results of extensive design experiments in which MAs were applied to optimize topologies and cross-sections of steel structural systems in tall buildings. Two classes of complex structural design problems were considered: conceptual and detailed design of a wind bracing system in a tall building and conceptual and detailed design of the entire steel structural system in a tall building. In these experiments, evolutionary algorithms (EAs) were employed to determine optimal configurations of structural members (conceptual design) while the sizing optimization (detailed design) was conducted using continuous/discrete optimization algorithm implemented in SODA (Grierson, 1989). The impact of all major evolutionary computation (EC) parameters on the performance of design processes was investigated. The following parameters were tested: the type of an evolutionary algorithm, parent and offspring population sizes, the type of the generational model, crossover and mutation rates, and the length of design processes (number of fitness evaluations).
The remainder of the paper is organized as follows. First, a brief description of MAs is provided. Next, experimental settings used in the computational experiments are presented. Furthermore, results of extensive computational studies are reported and grouped with respect to experimental parameters being tested. Finally, research conclusions are provided, including recommendations for optimal experimental settings for structural design experiments utilizing MAs as well as directions of future research.