基于随机森林回归的围岩应力插值方法  

An interpolation method of surrounding rock stress based on random forest regression

作　　者：王苏健 贾澎涛[2] 金声尧 WANG Sujian;JIA Pengtao;JIN Shengyao(Shaanxi Coal and Chemical Technology Institute Co.,Ltd.,Xi’an 710065,China;College of Computer Science and Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)

机构地区：[1]陕西煤业化工技术研究院有限责任公司,陕西西安710100 [2]西安科技大学计算机科学与技术学院,陕西西安710054

出　　处：《西安科技大学学报》2021年第2期274-281,共8页Journal of Xi’an University of Science and Technology

基　　金：国家重点研究发展计划(2018YFC0808303);国家自然科学基金资助项目(51974236)。

摘　　要：为解决矿井围岩应力监测数据缺失的问题,基于集成学习理论,提出了一种基于随机森林回归预测的围岩应力监测数据插值方法。首先,基于缺失前的历史数据和随机森林回归预测方法,建立围岩应力数据插值模型(IRER)。其次,以无缺失的围岩应力时间序列数据为样本集,构建不同缺失情况的数据集,作为实验用数据。最后,在不同缺失值情况下,选择均值插值、中值插值、线性插值、最邻近插值、Zero阶梯插值、3次B样条插值、拉格朗日插值7种插值方法作为实验对比插值方法,验证围岩应力数据插值模型。结果表明,均值插值、中值插值和拉格朗日插值方法效果较差,尤其是拉格朗日插值方法;随着缺失值个数的增大,拉格朗日插值方法的误差成倍增大;线性插值、最邻近插值、Zero阶梯插值、3次B样条插值适用于缺失值较少的情况;IRER方法在不同缺失值情况下,均取得了较好的插值效果,且随着缺失值数量的增加,这种优势尤为明显。In order to solve the problem of lack of monitoring data of mine surrounding rock stress,a new interpolation method is proposed based on random forest regression prediction algorithm of ensemble learning.Firstly,with the historical data and the random forest regression prediction algorithm in view,an interpolation model of surrounding rock stress is established.Secondly,the time series data of surrounding rock stress without missing is used as a sample set,and data sets with different missing conditions are constructed as the experimental data.Finally,in the case of different missing values,seven interpolation methods,including mean interpolation,median interpolation,linear interpolation,nearest neighbor interpolation,zero step interpolation,cubic B-spline interpolation and Lagrangian interpolation,were selected as experimental comparison interpolation methods to verify the interpolation model of surrounding rock stress data.The experimental results show that the mean interpolation,median interpolation and Lagrange interpolation are ineffective,especially the Lagrange interpolation method.With the number of missing values increasing,the error of Lagrange interpolation method increases sharply.Linear interpolation,nearest neighbor interpolation,zero step interpolation and cubic B-spline interpolation are suitable for the cases with fewer missing values.The random forest regression interpolation method achieves better effects under different missing values,and with the increasing of the number of missing values,this advantage is particularly obvious.

关 键 词：数据缺失 插值 集成学习 随机森林回归 预测 

分 类 号：X948[环境科学与工程—安全科学]