【摘要】 传统的结构或设备易损性分析方法需要提供大量的数值模拟样本，这对于规模庞大的核电结构并不适用。为此，研究核电设备高效易损性分析方法，首先，基于拉丁立方法构建随机地震动-土-结构样本，采用高效时域SSI分区并行计算方法得到部分样本模型的地震响应；然后，采用有限的数值模拟结果训练人工神经网络模型（ANN），通过量化ANN预测误差和精确度指标，采用自适应算法进行后续数值模拟和ANN训练，直至满足精确度阈值要求。该方法可以优化计算样本的选择，控制数值模拟的样本数量，提高易损性分析的计算效率。此外，将ANN不确定性整合到易损性曲线计算公式中，分别基于对数正态假定的回归法和蒙特卡洛（MC）增量法对某核电设备进行了易损性分析，并验证了ANN不确定性量化方法的正确性。更多还原

【Abstract】 Traditional methods of structural or equipment fragility analysis require a large number of numerical simulation samples,which is not applicable to large scale nuclear power structures. Therefore, an efficient technique for the fragility analysis of nuclear power equipment is developed in this paper. Firstly, the random ground motion-soil-structure samples are constructed based on the Latin method, and an efficient time-domain SSI partitioned parallel calculation method is used to obtain the seismic response of partial samples. Then, the neural network model(ANN) is trained with limited numerical simulation results, and by quantifying the prediction error and accuracy index of the ANN. An adaptive algorithm is used for subsequent numerical simulation and ANN training until the accuracy threshold requirements are all met. This technique can optimize the selection of calculation samples, control the number of samples for numerical simulation, and improve the calculation efficiency of fragility analysis. In addition, this paper integrates the ANN uncertainty into the calculation formula of the fragility curve, and conducts the fragility analysis of a nuclear power equipment based on the logarithmic regression method and the Mento Carlo incremental method, which verifies the ANN uncertainty.更多还原