基于Bi-LSTM神经网络的高层建筑塔吊安全预测方法  

Research on Safety Prediction Method of Tower Crane for High-rise Buildings Based on Bi-LSTM Neural Network

作　　者：李景新 杨国庆 徐增 LI Jingxin;YANG Guoqing;XU Zeng(School of Control and Mechanical Engineering,TCU,Tianjin 300384,China)

机构地区：[1]天津城建大学控制与机械工程学院,天津300384

出　　处：《天津城建大学学报》2025年第1期61-68,共8页Journal of Tianjin Chengjian University

基　　金：天津市研究生科研创新项目(2022SKYZ324)。

摘　　要：智慧工地的高层建筑塔吊安全是在建筑行业亟待解决的关键问题之一,塔身倾斜度是塔吊运动控制中的一个重要监测指标,为解决塔吊倾角预测精度不高问题,提出了残差学习(Res-Net)-双向长短期记忆神经网络(Bi-directional Long Short-Term Memory)模型预测高层建筑塔吊塔身各段倾角的方法.以分段监测的塔身倾角为输入,对塔吊塔身各段倾角实时监测预测.采用鲸鱼算法对模型进行优化,以最小化Res-Bi-LSTM网络的均方根误差为目标,寻找最优超参数,使得网络的误差最小.最终实现对塔身各段倾角的有效预测.实验结果分析提出的模型均方根误差(RMSE)降低到0.8%,模型的拟合优度达到94.96%,均优于对比实验的RNN、Bi-LSTM模型.本文所提出的模型具有更高的预测精度.The safety of tower cranes for high-rise buildings at smart construction sites is one of the key problems to be solved in the construction industry,and the inclination of the tower body is an important monitoring index in the tower crane motion control.To address the issue of low inclination prediction accuracy of tower cranes,this paper presents a residual learning-Bi-directional Long Short-Term Memory neural network model to predict the tilt angles of tower cranes in tall buildings.The real-time monitoring and prediction of each dip angle of the tower crane was carried out based on the input of the segmental monitoring.The whale algorithm was used to optimize the model,aiming to minimize the RMSE of Res-Bi-LSTM network,and to find the optimal hyperparameters to minimize the error of the network.It could ultimately realize the effective prediction of the inclination of each section of the tower.The root mean square error(RMSE)of the proposed model is reduced to 0.8%,and the goodness of fit of the model reaches 94.96%,both of which are better than the RNN and Bi-LSTM models of the comparison experiment.The model proposed in this paper has higher prediction accuracy.

关 键 词：高层建筑塔吊 分段监控系统 时间预测序列 鲸鱼优化算法 Bi-LSTM 

分 类 号：TU974[建筑科学—建筑技术科学]