基于机器学习LSTM方法的大面积高填土沉降预测  

Settlement prediction of large-area high filling soil based on machine learning method with LSTM

作　　者：王启贵 吴忠明 陈兴红 Wang Qigui;Wu Zhongming;Chen Xinghong(Zhejiang Huadong Surveying and Engineering Safely Co.,Ltd,Hangzhou 310014,China)

机构地区：[1]浙江华东测绘与工程安全技术有限公司,杭州310014

出　　处：《工程勘察》2022年第8期41-45,共5页Geotechnical Investigation & Surveying

摘　　要：大面积高填土一般具有区域地质条件差异性大,工程参数取值变异性大的特点,传统的地面沉降预测方法和以有限元为代表的数值分析方法都难以保证大面积高填土长期沉降预测的准确性。因此,本研究引入机器学习的方法,采用长短时记忆(LSTM)网络建立了云南省巧家县移民安置用地大面积高填方沉降预测模型。基于65组监测点位的沉降监测据,前60%的实测沉降数据作为LSTM模型训练样本,预测后40%的沉降数据。结果表明,本文提出的LSTM模型预测值与实际测量值之间有着较好的一致性,85%的监测点预测误差在20%以内,证明了本文提出的LSTM模型能够应用于大面积高填方沉降预测。Large-area high-fill soil has the characteristics of the large differences with the regional geological conditions and large variability in parameter values. The traditional surface subsidence prediction methods and numerical analysis methods represented by finite element are difficult to ensure the accuracy of prediction. In this paper, a long short-term memory(LSTM) network is proposed to establish a high-fill settlement prediction model. Based on a series set of settlement data from 65 groups of monitoring points, the first 60% of the settlement data is adopted as the training samples, where the last 40% of settlement data is used as the comparison purpose. The study results show that the predicted settlement values agree well with the measured values, the prediction errors of 85% settlement data fall within 20%, which confirms that the LSTM network can be used to predict the high-fill settlement accurately.

关 键 词：高填土 沉降预测 机器学习 长短时记忆网络 

分 类 号：TU449[建筑科学—岩土工程]