Steel plate shear wall optimization with metaheuristics an‎d neural network

12/22/2025 12:00:00 AM

Steel plate shear walls, serving as an efficient system for seismic resistance in buildings, frequently encounter challenges such as computational complexity in optimal design. This research introduces a hybrid approach aimed at minimizing the weight of these walls while complying with code regulations, employing artificial neural networks as surrogate models to approximate structural responses. First, a dataset comprising 40,000 samples was generated using OpenSeesPy software an‎d Latin hypercube sampling, connecting inputs like plate thicknesses an‎d beam-column sections to outputs including story drifts, average plate stresses, an‎d the structureʹs fundamental period. Subsequently, the neural network hyperparameters were tuned with five metaheuristic algorithms (differential evolution, particle swarm optimization, enhanced colliding bodies optimization, Jaya, an‎d artificial bee colony) to achieve a high-accuracy model. The results demonstrated that particle swarm optimization performed best in network tuning, with the surrogate model delivering a coefficient of determination exceeding 0.95 an‎d relative errors below 6% in predictions. Furthermore, during the weight optimization phase for a six-story shear wall, the enhanced colliding bodies optimization algorithm proposed a design weighing 45,823 pounds, which reduced computational time by approximately five times compared to direct analyses. These findings validate the practical applicability of this hybrid approach in seismic design.

: ديوار برشي فولادي , مدل جانشين , شبكه‌هاي عصبي مصنوعي , الگوريتم‌هاي فراكاوشي

Steel plate shear wall , surrogate model , artificial neural networks , metaheuristic algorithms

Sajjad Mansouri

Ali kaveh