基于双线性变化与多学习机器算法的矿山下沉估算比选模型

Comparative Model of Mining Subsidence Prediction Based on Bilinear Transformation and Machine Learning Algorithms

研究矿山地表最大下沉值的估算方法具有重要的安全生产意义,该研究的核心及关键问题在于提高估算的精度。本研究基于岩移数据决策、双线性变化及多种学习机器算法,对比筛选估算效果最优的矿山下沉估算方法。首先基于岩移数据决策方法,确定了采厚、倾角、平均采深、走向长度、倾向长度和覆岩岩性为最大下沉的影响参数;随后基于双线性变化的反推算模型,将倾角项转化为与其他参数相同时域数据以求提高估算精度;最后建立了C&RT、CHAID、Boosting Tree、Random Forest、BPNN和SVR等6种学习机器算法的估算模型。通过实例分析,CHAID和Random Forest方法返回了最差的估算结果,Boosting Tree和C&RT方法的估算结果会出现局部大残差值,BPNN估算时间数倍于其他方法,而SVR模型具有易于操作、耗时较短、精度较高的特点。故本研究认为SVR方法是一种高效可靠的最大下沉估算方法。

Accurate prediction of mine maximal surface subsidence is crucial for the safety of production,and the key issue is how to improve the accuracy of prediction.Based on decision-making method of strata movement data,bilinear transformation and machine learning algorithms,the optimal prediction of mining subsidence can be selected by comparison and analysis.Firstly,the parameters including mining thickness,dip angle,average mining depth,strike length,dip length and over-burden lithology are selected as relevant influence parameters to the maximum mining subsidence based on the decision-making method of strata movement data; secondly,the dip angle is converted into the standardized continuous parameter with the same time domain with other parameters based on the inversion calculation model based on bilinear transformation; finally,the prediction models of the machine learning algorithms such as CRT,CHAID,Boosting Tree,Random Forest,BPNN,and SVR are established.The performance evaluation results show that the worst prediction results are provided by CHAID and Random Forest,certain data points with large errors are produced by Boosting tree and CRT,the time consuming of BPNN is larger than others,SVR with the characteristics of high prediction accuracy and less computation cost.Therefore,SVR is the reliable and efficient one of mining subsidence prediction.