隧道围岩变形的非线性自回归时间序列预测方法研究

Research on nonlinear auto regressive time series method for predicting deformation of surrounding rock in tunnel

针对传统时间序列预测模型的单一线性和忽略施工过程影响的静态局限性,提出非线性自回归(包括NARNN与NARXNN)时间序列预测模型.该模型通过引入动态施工影响因子作为附加的外部输入,同时结合模型本身的反馈结构和延迟单元,在结构和动态特性上更加符合实际系统,可以非线性动态地考虑隧道施工全过程.运用该模型对史家山2号隧道施工过程中的围岩水平收敛和地表变形进行预测.结果表明:1)非线性自回归预测模型比传统的ARMA预测模型的预测精度高、适应性好;2)通过多次预测并对结果取平均值,可以保证非线性自回归预测模型预测结果的预测精度和稳健性;3)通过优化动态施工影响因子的取值方法,可以进一步提高NARXNN时间序列预测模型的预测精度.

Because traditional time series prediction models have the characteristics of single linear and static limitation due to the ignorance of the impacts of construction process,nonlinear auto regressive（including NARNN and NARXNN）time series prediction models are proposed in this paper.The models have their own feedback architectural and delay units,whose structural and dynamic properties are more coincide with the actual tunnel projects.Meanwhile,in order to nonlinearly and dynamically represent the tunneling process,dynamic construction impact factors,as apart of additional external inputs,are applied in this prediction model.Based on the nonlinear auto regressive time series prediction models,the transversal convergence and ground surface defo-rmation of Shijiashan 2nd tunnel are calculated.The comparison between the prediction results and the actual values shows that：1）Compared with the traditional ARMA time series prediction model,nonlinear auto regressive time series prediction models have a better adaptability and a higher precision.2）The prediction precision and the robustness of the nonlinear auto regressive prediction models can be improved by multiple calculations and taking the average.3）The prediction precision of NARXNN time series prediction models can be improved by the optimizing the value of dynamic construction impact factors.