Artificial Neural Network (ANN) model was developed as a reliable modeling method for simulating and predicting the ultimate moment capacities of castellated steel beams. The training and testing data for neural networks are obtained using Finite Element Analysis (FEA). For this purpose, a series of nonlinear finite element analyses have been carried out to simulate the distortional buckling behavior of castellated steel beams, and the effects of nine independent parameters on the lateral-distortional buckling mode, have been investigated. Moreover, unlike the existing design codes, the ANN model considers the effects of web distortion on the ultimate buckling strength of beams. Then, a new formula based on ANNs has been proposed to predict the ultimate moment capacities of castellated steel beams subjected to lateral-distortional buckling. The attempt was done to evaluate a practical formula considering all parameters which may affect the distortional capacity of castellated steel beams. Then, a sensitivity analysis using Garson’s algorithm has been developed to determine the importance of each input parameter. Finally, a comparison has been made between the proposed formula and the predictions obtained from AS4100, EC3, and AISC codes. It is shown that the proposed formula is more accurate than these design codes.

人工神经网络（ANN）模型是一种可靠的建模方法，用于模拟和预测带齿钢梁的极限弯矩能力。使用有限元分析（FEA）获得神经网络的训练和测试数据。为此，已经进行了一系列非线性有限元分析，以模拟带齿钢梁的变形屈曲行为，并研究了九个独立参数对横向变形屈曲模式的影响。此外，与现有的设计规范不同，ANN模型考虑了腹板变形对梁的极限屈曲强度的影响。然后，提出了一种基于人工神经网络的新公式来预测钢梁受侧向变形屈曲的极限弯矩承载力。尝试评估考虑所有可能影响齿形钢梁变形能力的参数的实用公式。然后，开发了使用Garson算法的灵敏度分析，以确定每个输入参数的重要性。最后，在提议的公式与从AS4100，EC3和AISC代码获得的预测之间进行了比较。结果表明，所提出的公式比这些设计规范更为准确。在提议的公式与从AS4100，EC3和AISC代码获得的预测之间进行了比较。结果表明，所提出的公式比这些设计规范更为准确。在提议的公式与从AS4100，EC3和AISC代码获得的预测之间进行了比较。结果表明，所提出的公式比这些设计规范更为准确。