基于遗传算法的基坑多指标变化量动态预测研究  

Research on Dynamic Prediction of Multi-Index Variables During Foundation Excavation Based on the Genetic Algorithm

作　　者：付新政 崔春雨 张乾青[3] 王思瑞 薛有泉 高鹏 FU Xinzheng;CUI Chunyu;ZHANG Qianqing;WANG Sirui;XUE Youquan;GAO Peng(Shandong Luqiao Group Co.,Ltd.,Jinan 250014,China;Shandong Jiaotong University,Jinan 250357,China;Institute of Geotechnical and Underground Engineering,Research Center,Shandong University,Jinan 250061,China)

机构地区：[1]山东省路桥集团有限公司,济南250014 [2]山东交通学院,济南250357 [3]山东大学岩土与地下工程研究院,济南250061

出　　处：《工业建筑》2024年第11期33-40,共8页Industrial Construction

基　　金：国家自然科学基金面上项目(52278358);2020年山东省工业和信息化厅项目(202150100118)。

摘　　要：针对现有机器学习预测模型训练效率低、算法单一、可能带来陷入局部极值、无法收敛等问题,结合反向传播(BP)人工神经网络、遗传算法和残差网络建立了深基坑多指标变化量的遗传算法-残差网络混合模型(GA-ResNN动态预测模型)和施工风险评价方法,研发了基坑施工风险智能预警平台。研究表明:提出的GA-ResNN动态预测模型较BP神经网络模型和GA-BP网络模型具有更好的预测精度,风险评估模型可实现定量预测和定性风险等级的评价。基坑施工风险智能预警平台通过输出预测曲线和预警阈值,可提升基坑工程的智能化管理和风险管控水平。A hybrid model of genetic algorithm combined with residual network(the GA-ResNN dynamic prediction model)for multi-index variables during deep foundation excavation and the construction risk assessment method were established by combining the back propagation(BP)artificial neural network,genetic algorithm(GA),and residual network(ResNN)to address the problems of low training efficiency of existing machine learning prediction models,the possibility of a single algorithm falling into local optima,and inability to converge.An intelligent early warning platform for excavation construction risks was developed.Research showed that the proposed GA-ResNN dynamic prediction model was of better prediction accuracy compared with the BP neural network model and GA-BP network model and could realize quantitative prediction and qualitative risk level evaluation.The intelligent warning platform for foundation excavation construction risks could present the prediction curve and warning threshold,which could improve the intelligent management and risk control level for engineering projects of foundation excavation.In view of the low training efficiency of the existing machine learning prediction models and the single algorithm that may lead to local extremum and convergence failure,BP artificial neural network,heuristic algorithm and residual network are used in this paper to establish the GA-ResNN dynamic prediction model for variation of multi-index and construction risk assessment method of deep foundation pit.Moreover,an intelligent early-warning platform for foundation pit construction risks is developed.The research results show that the GA-ResNN dynamic prediction model proposed in this paper has better prediction accuracy than BP artificial neural network model and GA BP network model,and the risk assessment model can realize quantitative prediction and qualitative risk grade evaluation.The intelligent early-warning platform for foundation pit construction risk can improve the intelligent management and risk control level

关 键 词：基坑 机器学习 启发式遗传算法 动态预测 风险评估 

分 类 号：TU753[建筑科学—建筑技术科学]