基于BP神经网络的有限特征参数砌体结构震害等级推演

详细的建筑结构特征参数是得到合理地震易损性分析结果的基础。本文给出了一种结合已有地震易损性分析成果,在具备有限特征参数的情况下,利用BP神经网络进行单体或群体结构震害等级推演的方法。以陕西省渭南市607栋设防砌体易损性评估结果为样本构建了一个3层BP神经网络模型,并对北京市海淀区近2万栋设防砌体不同地震烈度下的可能破坏状态进行推演,结果能够反映区域本地化特征,也与抗震设计目标和震害案例相符。该方法适用于少量特征参数下单体或群体结构的快速震害等级推演,可为相似烈度地区的建筑结构风险评估提供参考。

BP neural network prediction of seismic damage level of masonry structures with finite characteristic parameters

Detailed structural parameters are the basis of reasonable seismic vulnerability analysis. A method is proposed to predict the seismic damage level of structures based on BP neural network with finite characteristic parameters. After seismic vulnerability analysis of 607 fortification masonry structures in Weinan city,training samples were selected for the construction of a three-layer BP neural network,and the possible failure states of nearly 20000 masonry structures in Haidian District of Beijing have been predicted with the trained model. The results can reflect the localization characteristics of building structures,and also conform to the seismic design objectives and earthquake damage cases. The method can provide a reference for the earthquake damage prediction and risk prevention of building structures in similar intensity areas.