基于BP神经网络模型的村镇砖砌体结构震害预测研究
|
摘要
砖砌体结构是村镇地区一种量大面广的结构形式,其抗震性能薄弱,在地震中极易出现脆性破坏。本文尝试应用基于L-M算法的BP神经网络方法,利用它强大的非线性映射功能,建立起村镇地区砖砌体结构震害影响因素与破坏状态等级之间关系。设计出一个9-6-5的三层神经网络模型,根据实地调查,筛选出影响房屋震害的9个主要因素,如层数、层高、砌筑方式、砖墙面积率等作为神经网络的输入参数,输出参数为房屋5种破坏状态。选择2008汶川地震后四川、陕西、甘肃等地的震害实例作为学习样本对所构建的神经网络模型进行训练。训练结果表明,该模型对已训练数据有很好的适应性,但如果要将其用于单个或群体建筑的易损性分析,并取得较精确的预测结果,还需积累足够多的训练样本,并进行大量的网络试验工作。
Brick masonry structure is a widely-used structure type in the village.It has low seismic performance,and is prone to fragile damage in earthquake.This paper attempts to apply BP artificial neural networks(ANN),which has stronger nonlinear mapping function,to establish a relationship between the structural earthquake damage factors and the destruction state.A three-layer BP network model is designed,in which nine main damage factors,including the floor number,the height of floor,the masonry methods,the area ratio of brick wall are the input parameters,five destruction states are the output parameters.The earthquake examples in Sichuan,Shaanxi,Gansu after 2008 Wenchuan earthquake are chosen to be the learning samples for the established artificial neural networks.The training results indicate that the model has very successful adaptation for the training data,but if the model is used to analyze the seismic fragility of individual building or building group,and is expected to have accurate prediction results,more study samples and mass training work are still needed.
Brick masonry structure is a widely-used structure type in the village.It has low seismic performance,and is prone to fragile damage in earthquake.This paper attempts to apply BP artificial neural networks(ANN),which has stronger nonlinear mapping function,to establish a relationship between the structural earthquake damage factors and the destruction state.A three-layer BP network model is designed,in which nine main damage factors,including the floor number,the height of floor,the masonry methods,the area ratio of brick wall are the input parameters,five destruction states are the output parameters.The earthquake examples in Sichuan,Shaanxi,Gansu after 2008 Wenchuan earthquake are chosen to be the learning samples for the established artificial neural networks.The training results indicate that the model has very successful adaptation for the training data,but if the model is used to analyze the seismic fragility of individual building or building group,and is expected to have accurate prediction results,more study samples and mass training work are still needed.
引文
[1]袁雪霞,赵艳林.多层砖房震害预测的人工神经网络方法[J].广西大学学报:自然科学版,2000,25(3):234-237.
[2]侯媛彬.神经网络[M].西安:西安电子科技大学出版社,2007:35.
[3]周志华.神经网络及其应用[M].北京:清华大学出版社,2007:35.
[4]陆卫东,王继仁,单亚飞,等.基于L-MBP神经网络的煤自然发火期预测[J].辽宁工程技术大学学报,2005,25(6):815-818.
[5]吕俊,张兴华.几种快速BP算法的比较研究[J].现代电子技术,2003,167(24):96-99.
[6]陈万海,王婷,苗延辉.基于LMBP神经网络的非线性畸变图形校正研究[J].哈尔滨工程大学学报,2006,27:173-176.
[7]汤皓,陈国兴,李方明.基于BP神经网络模型的多层砖房震害预测方法[J].地震工程与工程振动,2006,26(4):141-146.
[8]汤皓,陈国兴,李方明.灰关联与人工神经网络在建筑震害预测中的应用[J].地震工程与工程振动,2006,26(3):58-59.
[9]刘本玉,叶燎原,江见鲸.用模糊人工神经网络方法预测多层砖房震害[J].清华大学学报:自然科学版,2002,42(6):843-846.
[10]缪升,刘本玉,叶燎原.多层砖房震害预测的人工神经网络方法及实践[J].世界地震工程,2000,16(1):20-25.
[2]侯媛彬.神经网络[M].西安:西安电子科技大学出版社,2007:35.
[3]周志华.神经网络及其应用[M].北京:清华大学出版社,2007:35.
[4]陆卫东,王继仁,单亚飞,等.基于L-MBP神经网络的煤自然发火期预测[J].辽宁工程技术大学学报,2005,25(6):815-818.
[5]吕俊,张兴华.几种快速BP算法的比较研究[J].现代电子技术,2003,167(24):96-99.
[6]陈万海,王婷,苗延辉.基于LMBP神经网络的非线性畸变图形校正研究[J].哈尔滨工程大学学报,2006,27:173-176.
[7]汤皓,陈国兴,李方明.基于BP神经网络模型的多层砖房震害预测方法[J].地震工程与工程振动,2006,26(4):141-146.
[8]汤皓,陈国兴,李方明.灰关联与人工神经网络在建筑震害预测中的应用[J].地震工程与工程振动,2006,26(3):58-59.
[9]刘本玉,叶燎原,江见鲸.用模糊人工神经网络方法预测多层砖房震害[J].清华大学学报:自然科学版,2002,42(6):843-846.
[10]缪升,刘本玉,叶燎原.多层砖房震害预测的人工神经网络方法及实践[J].世界地震工程,2000,16(1):20-25.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心