深基坑施工时地表沉降预测的时序-投影寻踪回归模型

为了保证施工的正常进行、实现信息化施工，必须对基坑实际监测数据进行分析与预测。现有的趋势时间序列分析方法很难满足实际施工中高度非线性问题的拟合，预测误差较大。基于这点考虑，以天津某工程基坑施工地表沉降观测序列为例，在对原始数据进行分析的基础上，提出既可以考虑趋势时间序列，又具有高度非线性拟合性能的时序-投影寻踪回归模型。首先，通过比较分析几种时间序列方法的逼近误差和预测误差，寻求出一种逼近较好的时间序列预测方法。然后，将预测得到的时间序列和观测数据相结合，采用投影寻踪回归方法拟合。应用结果表明，该模型逼近性能良好，预测误差小，可为深基坑位移沉降的动态预测提供一条较好的途径，对基坑动态设计与信息化施工等方面具有重要的参考价值。

Time series-projection pursuit regression model for predicting surface settlement during pit excavation

To make the construction safe and implement the informative construction, it is important to analyze and predict the practical monitoring data during pit excavation. Time series analysis of existing trends is very difficult to meet the actual construction of highly nonlinear problems of fitting; and forecasting error is big. Based on this consideration, taking the practical monitoring data of the surface settlement from a pit construction in Tianjing for example, the time series-projection pursuit regression model is presented. After analyzing the raw data, the model can be considered both time series and the highly nonlinear fitting performance. First, comparative analysis of the approximation error and the forecast error of several time series methods, we find an approximation of good time series forecasting methods. The predicted and observed time series data combined with projection pursuit regression fitting. Application results show that the approximation performance of proposed model is good; and its prediction error is small. The methods presented herein are important reference to the dynamic design and informative construction for pit engineering.