The rack column is one of the essential elements in the pallet rack system. However, due to its distinctive perforation feature, it is challenging to analyze its stability using traditional theories for cold-formed steel structures. In this paper, we are interested in the comparison analysis of strength prediction on the perforated columns using finite element method (FEM), regression analysis (RA) and artificial neural network (ANN) methods respectively. First, a refined finite element (FE) model considering the perforation and nonlinearity behavior was generated and calibrated against the experimental results. Subsequently, the validated FE model was used to perform the parametric analysis for the different holes in columns. Given experimental and simulated data, a regression model with an equivalent thickness was proposed for the design strength prediction of thin-walled steel perforated sections. For comparison of the RA model, two powerful tools such as the FEM and ANN are also employed to predict the design strength of different perforated sections. Four indicators were used to assess the accuracy and generalization performance of the three models, including the root mean square error, the mean absolute percentage error, the correlation coefficient and the mean relative percentage. The obtained results show that although they both have good consistency, FEM still slightly outperforms the other two models. Since the values calculated from ANN and regression models are usually smaller than the experimental data, they are reasonably recommended as effective and safer design tools than FEM models from the perspective of engineering applications.

货架栏是托盘货架系统中的基本要素之一。然而，由于其独特的穿孔特征，使用传统理论对冷弯钢结构进行稳定性分析具有挑战性。在本文中，我们对分别使用有限元方法（FEM），回归分析（RA）和人工神经网络（ANN）方法的多孔柱强度预测的比较分析感兴趣。首先，生成考虑了穿孔和非线性行为的精制有限元（FE）模型，并根据实验结果进行了校准。随后，使用经过验证的有限元模型对列中的不同孔进行参数分析。给定实验和模拟数据，提出了等效厚度的回归模型，用于薄壁钢穿孔截面的设计强度预测。为了比较RA模型，还使用了FEM和ANN等两个功能强大的工具来预测不同穿孔部分的设计强度。四个指标用于评估三个模型的准确性和泛化性能，包括均方根误差，平均绝对百分比误差，相关系数和平均相对百分比。所得结果表明，尽管两者都具有良好的一致性，但FEM仍然略胜于其他两个模型。由于根据ANN和回归模型计算得出的值通常小于实验数据，