---

This paper combines particle swarm optimization, grid search method and univariate method as a general optimization approach for any type of problems emphasizing on optimum design of steel frame structures. The new algorithm is denoted as the GSU-PSO. This method attempts to decrease the search space and only searches the space near the optimum point. To achieve this aim, the whole search space is divided into a series of grids by applying the grid search method. By using a method derived from the univariate method, the variables of the best particle change values. Finally, by considering an interval adjustment to the variables and generating particles randomly in new intervals, the particle swarm optimization allows us to swiftly find the optimum solution. This method causes converge to the optimum solution more rapidly and with less number of analyses involved. The proposed GSU-PSO algorithm is tested on several steel frames from the literature. The algorithm is implemented by interfacing MATLAB mathematical software and SAP2000 structural analysis code. The results indicated that this method has a higher convergence speed towards the optimal solution compared to the conventional and some well-known meta-heuristic algorithms. In comparison to the PSO algorithm, the proposed method required around 45% of the total number of analyses recorded and improved marginally the accuracy of solutions.

---

This paper proposes a GA-based reduced search space technique (GA-RSS) for the optimal design of steel moment frames. It tries to reduce the computation time by focusing the search around the boundaries of the constraints, using a ranking-based constraint handling to enhance the efficiency of the algorithm. This attempt to reduce the search space is due to the fact that in most optimization problems the optimal solution lies on or near the boundaries of the feasible region. All the analyses/optimization steps have been implemented in MATLAB and the method has been validated by optimizing three moment-frame benchmark problems. According to the results, the algorithm performs fit and needs relatively fewer analyses than other metaheuristic algorithms to reach a global optimum solution.