Determinants of Design with Multilayer Perceptron Neural Networks: A Comparison with Logistic Regression

This research focuses on harnessing artificial neural networks (ANNs) to enhance the design of steel structures. The design process encompasses various stages, including defining the building’s geometry, estimating loads, selecting an appropriate structural system, sizing components, and creating detailed plans. Optimizing the weight of these structures is vital for reducing costs, improving efficiency, and minimizing environmental impact. This study specifically investigates multilayer perceptron (MLP) neural networks to optimize steel structure design. It evaluates different ANN configurations with varying numbers of hidden layers and neurons to find the most effective arrangement. Additionally, the performance of MLP networks is compared to that of logistic regression. The results demonstrate that MLP networks deliver superior accuracy in optimizing the design of steel structures compared to logistic regression. The process of designing steel structures at an early stage can reduce the con-sumption of energy and raw materials before the production of the structures themselves begins. This is important from an economic point of view because some costs can be reduced during the design process. When designing steel structures, it is also possible to take into account changing conditions, such as the growing share of re-newable energy sources in the total energy balance in many countries.

本研究旨在利用人工神经网络（Artificial Neural Networks，ANNS）优化钢结构设计流程。钢结构设计涵盖多个核心环节，包括确定建筑几何形态、开展荷载估算、选取适宜的结构体系、完成构件截面选型以及绘制详细施工图纸。优化钢结构自重对于降低建造成本、提升结构能效以及减少环境影响均至关重要。本研究专门针对多层感知器（Multilayer Perceptron，MLP）神经网络展开，用于优化钢结构设计方案。研究中对比了不同隐藏层层数与神经元数量的人工神经网络架构，以探寻性能最优的网络配置方案。此外，还将多层感知器网络的性能与逻辑回归（Logistic Regression）模型进行了对比。实验结果表明，相较于逻辑回归模型，多层感知器网络在钢结构设计优化任务中具备更优异的预测精度。在钢结构生产启动前的早期设计阶段，即可降低能源与原材料的消耗。从经济视角来看，此举意义重大，因为在设计阶段即可削减部分不必要的建造成本。此外，在钢结构设计过程中，还可兼顾动态变化的外部条件，例如诸多国家的能源总平衡中可再生能源占比持续提升这一现状。