建筑结构学报2025，Vol.46Issue(4)：177-187,209,12.DOI:10.14006/j.jzjgxb.2024.0247
古建筑砌体结构裂缝宽度预测及监测数据异常识别方法
Prediction of crack width in ancient masonry structures and methods for anomaly detection in monitoring data
摘要
Abstract
Predicting the changes of cracks in ancient buildings is essential for achieving preventive conservation.However,the complex characteristics of the ancient structures pose challenges to the prediction accuracy.Therefore,an EMD-LSTM-SVR prediction model was proposed.This model integrated the data simplification ability of empirical mode decomposition(EMD)algorithm,the high-accuracy prediction characteristic of long short-term memory network(LSTM)model,and the high-efficiency advantage of support vector regression(SVR)model.Utilizing the crack width monitoring data sequences from a Tibetan-style ancient masonry wall,the parameters of LSTM and SVR models were optimized through grid search algorithm in conjunction with intrinsic mode functions extracted by the EMD algorithm to achieve optimal prediction outcomes.Based on both predicted and actual values,statistical process control(SPC)was then employed to detect anomalies in the monitoring data.The research results indicate that,compared with single or combined models such as LSTM,SVR,EMD-LSTM,and LSTM-SVR,the EMD-LSTM-SVR prediction model demonstrates higher accuracy and efficiency,with a mean absolute error of only 0.010 4.The SVR model exhibits the largest prediction deviation.An SPC anomaly detection analysis was conducted on the differences between the prediction results and the actual measured data,revealing that the proportions of all eight tested anomaly phenomena are less than 2％.关键词
古建筑砌体结构/裂缝宽度/经验模态分解算法/长短时记忆网络模型/支持向量回归模型/统计过程控制Key words
masonry structure of history building/crack width/empirical mode decomposition algorithm/long short-term memory network model/support vector regression model/statistical process control分类
建筑与水利引用本文复制引用
杨娜,王烁,付颖煜..古建筑砌体结构裂缝宽度预测及监测数据异常识别方法[J].建筑结构学报,2025,46(4):177-187,209,12.基金项目
中央高校基本科研业务费专项资金资助(2021JBZ110),国家自然科学基金面上项目(52478119,51878034),高等学校学科创新引智计划(B13002). （2021JBZ110）
