摘要
砖混建筑物在地震中破坏严重损失巨大,快速评估地震作用下城市砖混建筑物的破坏风险实现建筑抗震韧性至关重要。传统神经网络预测方法在建筑震害分析过程中存在易陷入局部最优和收敛效率低等问题。为此,本文提出一种耦合SMIV(Spearman-Mean Impact Value)和PSO-LMBP(Particle Swarm Optimization-Levenberg Marquard Back Propagation)的砖混结构群集成震害预测方法。首先,利用SMIV方法进行震害因子筛选降低数据维数;其次,建立了耦合PSO(Particle Swarm Optimization)和LM(Levenberg Marquard)算法的PSO-LMBP神经网络的震害预测模型,通过PSO算法将得到的一组全局最优解作为BP网络的初始权值和阈值,再利用LM算法对BP神经网络进行优化训练;最后,从整体样本的预测精度、拟合效果以及运行速度上进行对比分析,交叉验证结果表明提出的SMIV-PSO-LMBP模型震害预测效果显著。同时,以广州地区为例,应用本文提出的方法进行了区域砖混结构群震害预测,预测结果与华南地区砖混建筑实际计算统计得到的震害矩阵对比,误差较小。综上所述,本文提出的SMIV-PSO-LMBP预测方法能够较好、较快地评估出区域砖混建筑物的破坏风险,为政府震后进行精准救灾提供一定的借鉴。
Brick-concrete buildings suffer from severe damage and huge losses in earthquakes.It is very important to quickly assess the damage risk of urban brick-concrete buildings under the action of earthquakes to achieve seismic resilience of buildings.In the process of building earthquake damage analysis,traditional neural network prediction methods are prone to fall into local optimum and have low convergence efficiency.To this end,This paper proposes a coupling SMIV(Spearman-Mean Impact Value)and PSO-LMBP(Particle Swarm Optimization-Levenberg Marquard Back Propagation)method for earthquake damage prediction of brick-concrete structures.Firstly,SMIV method is used to screen earthquake damage factors to reduce the data dimension.Secondly,the earthquake damage prediction model of PSO-LMBP neural network coupled with PSO(Particle Swarm Optimization)algorithm and LM(Levenberg Marquard)algorithm was established.Through PSO algorithm,a set of global optimal solutions were used as the initial weights and thresholds of BP network.Then LM algorithm is used to optimize BP neural network training.Finally,the prediction accuracy,fitting effect and running speed of the whole sample are compared and analyzed.The cross-validation results show that the proposed SMIV-PSO-LMBP model has a significant effect on earthquake damage prediction.At the same time,taking Guangzhou as an example,the method proposed in this paper is used to predict the regional group earthquake damage of brick-concrete structures.The predicted results are compared with the earthquake damage matrix of brick-concrete buildings in South China,and the error is small.To sum up,the SMIV-PSO-LMBP prediction method proposed in this paper can better and quickly assess the damage risk of regional brick and concrete buildings,and provide certain reference for the government to carry out accurate disaster relief after earthquakes.
作者
孙海
邢启航
姜慧
阮雪景
刘孟佳
SUN Hai;XING Qihang;JIANG Hui;RUAN Xuejing;LIU Mengjia(College of Engineering,Ocean University of China,Qingdao 266100,China;Institute of Marine Development of Ocean University of China,Qingdao 266100,China;Guangdong Earthquake Agency Key Laboratory of Earthquake Monitoring and Disaster Mitigation Technology,Guangzhou 510070,China;College of Civil Engineering&Architecture,Qingdao Agricultural University,Qingdao 266109,China)
出处
《世界地震工程》
北大核心
2023年第3期154-164,共11页
World Earthquake Engineering
基金
国家自然科学基金项目(U1901602-05,41906185,52071307)
山东省科技重大专项(2020CXGC010702)
基于地震风险评估的强震灾害情景构建及应用示范。
关键词
震害预测
LMBP网络
粒子群算法
SMIV算法
砖混结构
earthquake damage prediction
LMBP neural network
particle swarm optimization
SMIV algorithm
masonry structure
