四肢钢管混凝土格构柱抗震性能的神经网络优化预测  

Neural network optimization of seismic performance prediction for four-limbconcrete-filled steel tubular lattice columns

作　　者：陈娟 何俊杰 黄志 周旺保[4] 戚菁菁[2] CHEN Juan;HE Junjie;HUANG Zhi;ZHOU Wangbao;QI Jingjing(School of Information and Electrical Engineering,Hunan University of Science and Technology,Xiangtan 411201,China;School of Civil Engineering,Hunan University of Science and Technology,Xiangtan 411201,China;Hunan Engineering Research Center for Intelligently Prefabricate Passive House,Xiangtan 411201,China;School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]湖南科技大学信息与电气工程学院,湖南湘潭411201 [2]湖南科技大学土木工程学院,湖南湘潭411201 [3]湖南省智慧建造装配式被动房工程技术研究中心,湖南湘潭411201 [4]中南大学土木工程学院,湖南长沙410075

出　　处：《自然灾害学报》2024年第6期192-205,共14页Journal of Natural Disasters

基　　金：国家自然科学基金项目(52204210,51808213);湖南省自然科学基金项目(2023JJ30242);湖南省教育厅科学研究优秀青年项目(21B0452,20B214);湖南省教育厅科学研究重点项目(20A184)。

摘　　要：为了能够精确预测出四肢钢管混凝土格构柱在低周往复荷载作用下的荷载-位移滞回曲线,采用麻雀搜索优化算法优化Elman神经网络模型(SSA-Elman)对四肢钢管混凝土格构柱在低周往复荷载作用下的荷载-位移滞回曲线进行了预测。通过5个标准测试函数对麻雀搜索算法(sparrow search optimization algorithm,SSA)、粒子群优化算法(particle swarm optimization,PSO)和萤火虫算法(firefly algorithm,FA)进行了性能比较。对4根不同参数的四肢钢管混凝土格构柱进行了低周往复荷载试验,得到了其荷载-位移滞回曲线并进行了分析。采用荷载-位移滞回曲线数据对各模型进行了训练和预测并对误差结果进行了分析。基于传统的能量损伤模型和Park模型,计算SSA-Elman模型预测数据的损伤值和试验数据损伤值,并进行对比。结果表明:在低周往复荷载作用下,试件SCC1和试件SCC2的荷载-位移滞回曲线的形状为弓形,试件SCC3的荷载位移-滞回曲线为反S形,试件SCC4的荷载-位移滞回曲线更加接近于梭形;通过5个标准测试函数对比得到SSA算法搜索性能优于PSO和FA算法,具有较好的寻优精度、收敛速度和全局搜索能力;SSA-Elman模型对4个试件抗震性能预测的平均绝对百分比误差值均小于Elman、PSO-SVR和FA-BP模型,SSA算法能有效优化Elman模型的权值和阈值;SSA-Elman预测模型获得的损伤值与传统模型损伤值能够很好的吻合,验证了SSA-Elman模型用于预测四肢钢管混凝土格构柱在低周往复荷载作用下的荷载-位移滞回曲线是可行的。To predict the load-displacement hysteresis curves of four-limb concrete-filled steel tubular(CFST)lattice columns under low-cycle reciprocating loading tests,the research employs the sparrow search optimization algorithm(SSA)to optimize the Elman neural network model(SSA-Elman).The performance of SSA is compared with particle swarm optimization(PSO)and firefly algorithm(FA)using five standard test functions.Low-cycle reciprocating loading tests were conducted on four-limb CFST lattice columns to determine their damage patterns and load-displacement hysteresis curves.The obtained test results are utilized for training,prediction,and error analysis of each model.Damage values are calculated and compared between the predicted results of the SSAElman model and the test data based on traditional energy damage models and Park’s model.The results indicate that the load-displacement hysteresis curves of specimens SCC1 and SCC2 are arched,while specimen SCC3 exhibits an anti-S-shaped curve,and specimen SCC4 shows a shuttle-shaped curve.Comparisons among the five standard test functions reveal that the SSA algorithm outperforms PSO and FA algorithms in terms of search performance,optimization accuracy,convergence speed,and global search ability.The SSA-Elman model demonstrates lower mean absolute percentage error(MAPE)when predicting the seismic performance of the four specimens compared with Elman,PSO-SVR,and FA-BP.The SSA technique effectively optimizes the weights and thresholds of the Elman model.Moreover,the damage values predicted by the SSA-Elman model are consistent with those obtained from the traditional model,suggesting that the SSA-Elman model is suitable for predicting the loaddisplacement hysteresis curve of four-limb CFST lattice columns under low-cycle reciprocating loading tests.

关 键 词：格构柱 Elman模型 损伤预测 低周往复荷载 SSA算法 滞回曲线 

分 类 号：TU375.3[建筑科学—结构工程] TU352