深度学习在钢结构货架变形预测中的应用研究  

Application study of deep learning algorithm for steel structure shelves deformation prediction

作　　者：魏来 张雅晨 潘健 胡一清 Wei Lai;Zhang Yachen;Pan Jian;Hu Yiqing(Nanjing Yuanneng Power Engineering Co.,Ltd.,Nanjing Jiangsu 210000,China)

机构地区：[1]南京远能电力工程有限公司,江苏南京210000

出　　处：《山西建筑》2025年第2期28-32,43,共6页Shanxi Architecture

基　　金：“数字周岗”改造工程建设项目(编号:823140816)。

摘　　要：随着工业化和物流行业的发展,钢结构货架在仓储和物流系统中越来越重要,因此准确预测其变形至关重要。文章介绍了一种基于双向长短时记忆网络(BiLSTM)和注意力机制的预测算法,该算法利用时间序列数据,通过深度学习模型进行训练,能够更细致地分析和预测钢结构货架的变形。结合一个典型应用验证了模型性能,证实了其高稳健性和出色的预测精度。实验结果表明,该模型能够准确地预测钢结构货架的变形情况,其平均误差仅为0.15%~3.33%。这些结果表明了该算法在钢结构货架自动化监测领域的潜在应用前景,为其结构变形预测提供了一种可行的解决方案。With the growth of industrialization and the logistics industry,steel structure shelves have become increasingly vital in storage and logistics systems,making accurate deformation prediction essential.This paper presents a prediction algorithm based on Bidirectional Long Short-Term Memory Networks(BiLSTM)and Attention mechanisms.Utilizing time series data and training through deep learning models,this algorithm allows for detailed analysis and prediction of deformations in steel structure shelves.The model’s performance,validated by a typical application,shows high robustness and exceptional predictive accuracy.Experimental results indicate that the model can accurately predict deformations in steel structure shelves,with an average error of only 0.15%to 3.33%.These outcomes suggest potential applications of this algorithm in the automated monitoring of steel structure shelves,providing a viable solution for predicting structural deformations.

关 键 词：自动化监测 深度学习 时间序列数据 双向长短时记忆网络与注意力机制(BiLSTM-Attention) 

分 类 号：TU391[建筑科学—结构工程]