与短持时地震动相比，长持时地震动会加剧结构的损伤，增加结构的失效概率，因此有必要更充分地研究地震动持时特性对结构地震易损性分析结果的影响。该文提出了一种基于BP神经网络的地震易损性曲面分析方法，使用神经网络模型，综合考虑地震动强度和持时特性对结构地震需求的影响，并进行地震易损性分析，得到不同损伤水平下考虑地震动持时特性的结构易损性曲面。选用3个不同高度的钢筋混凝土框架结构为研究对象，分别选择具有长、短持时特性的2组地震动记录为输入，采用BP神经网络模型建立地震动强度指标与结构响应间的关系，在此基础上得到目标地震易损性曲面，并对该方法的有效性进行讨论。分析结果表明，研究建立的BP神经网络模型精度较高，依据该方法可得到可信的损伤概率分析结果。相比于传统方法，神经网络可以更为有效和准确地建立持时与结构损伤的相关关系，得到考虑持时特性的易损性分析结果。该文的方法亦可进一步拓展，将更多地震动特性纳入地震易损性分析过程，具有明确的应用前景。“,”Comparing with short-duration ground motions, long-duration ground motions may intensify the damage and increase the failure probability of structures. Therefore, it is necessary to thoroughly investigate the influence of ground motion duration characteristics on the seismic fragility analysis results. A seismic fragility surface analysis approach based on back propagation (BP) artificial neural networks was proposed. It can account for the effect of both ground motion intensity and duration. Seismic fragility analysis was conducted to get the fragility surfaces under different damage levels. Three reinforced concrete fame structures with different heights were taken as the study cases. Long- and short-duration record sets were selected as the inputs. BP neural network models were employed to build the relationship between the ground motion intensity measures and structural responses, and the seismic fragility surfaces of the investigated structures were obtained. The validity of the proposed approach was discussed. The analysis results show that the accuracy of the established BP neural network model is high. It indicates that the fragility analysis results by this approach is reliable. Comparing with the conventional procedures, the neural network is capable of building more effective correlation models between the ground motion duration and structural damage to obtain fragility analysis results that account for ground motion duration. This approach can be further expanded to include more ground motion characteristics into the program for seismic fragility analysis. It has a definite application prospect.