【作者】 吴琼；

【导师】 胡长明；

【作者基本信息】 西安建筑科技大学 ， 土木工程建造与管理， 2015， 硕士

【摘要】 地震是一种典型的自然灾害,其发生几率小但破坏力巨大,所造成的损失往往难以估量。我国处于环太平洋地震带与欧亚地震带之间,地震活动十分频繁。对于经济快速发展、人口较为密集的中国而言,地震威胁日益严重。所以,对地震直接经济损失的快速评估理论进行系统研究具有非常重要的理论意义和应用价值。本文采用易损性分析方法,基于对地震造成的经济损失的系统分析结果,将建筑物按结构类型及使用功能进行分类,建立了建筑物结构破坏、建筑物装饰、建筑物室内财产直接经济损失的快速评估模型,在此基础上重点对建筑物重置单价及建筑物破坏损失比进行了研究。运用SPSS软件对建筑物重置单价与年份之间的关系进行了一元回归分析,进而得到了拟合优度较高且能通过显著性检验的回归方程。根据对致灾及承灾两方面因素的定性分析,选择地震震级、震源深度、抗震设防烈度、设计基本地震加速度、建筑物重置单价、GDP等六个指标为输入变量,五种不同破坏状态下的建筑物结构破坏损失比为输出变量,构建了BP神经网络,调试并选择了LM算法进行神经网络的训练,并采用提前终止法来提高神经网络的泛化能力,最终得到了满足评估速度及精度要求的BP神经网络。与此同时,本文通过系统总结权威的研究成果,为建筑物装修损失及室内财产损失的评估提供了可靠的参考。论文得到的初步成果可为地震灾区直接经济损失的快速评估提供有效的方法,也可为制定相关应急预案及防灾减灾措施提供依据。更多还原

【Abstract】 As a typical natural disaster with lower occurrence and huger damage, earthquake often causes inestimable losses. Located between Circum-Pacific seismic belt and Eurasia seismic belt, China suffers from quite frequent seismic activities. And with the rapid economic development and gradually intensive personnel, there is a growing seismic threat for our country. Therefore, a systematic research on the rapid evaluation method of earthquake-caused direct economic loss takes on an important theoretic meaning and realistic values.Based on the analysis results of earthquake-caused economic loss, this paper employs the fragility analysis, classifies buildings according to structure type and occupancy class, models the direct economic loss from three aspects and mainly researches the unit replacement cost and the structural loss ratio. With the use of SPSS, a unitary linear recursive analysis of the relation between unit replacement cost and year is performed and regression equations which are with higher goodness of fit and approved by significant test are obtained. Through qualitative analysis of hazard factors and characteristics of hazard-affected bodies, earthquake magnitude, focal depth, seismic fortification intensity, basic design acceleration of ground motion, unit replacement cost and GDP are chosen as input variables, while structural loss ratios under the different damage states are output variables. A back propagation neural network(BPNN) is built and the Levenberg-Marquardt(LM) algorithm is used to train the network. Meanwhile, the generalization ability is improved by the early termination algorithm. The training results meet the requirements of speed and accuracy. Through systematic summary of authoritative research, this paper also provides convincing reference to the evaluation of decoration loss and building contents loss.The preliminary results can offer an effective method for the rapid evaluation of the earthquake-caused direct economic loss and provide basis for contingency plans and hazard reduction measures.更多还原