目前岩爆预测的真实训练数据量小、数据存在缺失，为了更加准确地预测岩爆等级，提出了一种基于链式随机森林多重插补（MICE_RF）算法的组合赋权—极限随机树（ET）预测模型。首先，在选取岩爆灾害主要评判指标的基础上，采用MICE_RF算法插补缺失数据；然后，由改进层次分析法（IAHP）和基于指标相关性的权重确定方法（CRITIC）确定指标主、客观权重，并引入权向量距离概念对指标组合赋权；最后，将插补和赋权后数据集采用ET算法，构建岩爆等级预测模型。利用国内外工程实例数据进行20次随机抽样试验，并与其他模型进行对比分析。结果表明：MICE_RF插补后可显著提高岩爆模型预测效果；改进AHP-CRITIC法较改进前更具优势，该模型平均预测准确率为93.10%，各比较指标结果均优于对比模型，预测结果更稳定。

As a kind of dynamic instability geological disaster with strong abruptness and randomness，rockburst poses a great threat to the safety of personnel，equipment and buildings.Timely and accurate prediction of rockburst grade has become a hot issue in the field of underground engineering.At present，the amount of real training data of rockburst prediction is small and the data is missing.In order to predict the rockburst grade more accurately，a combined weighting-extremely randomized trees（ET） prediction model based on chain random forest multiple interpolation（MICE_RF） algorithm was proposed.According to the characteristics and causes of rockburst，six evaluation indexes including maximum shear stress，uniaxial compressive strength，uniaxial tensile strength，stress coefficient，brittleness coefficient and elastic energy index were selected to form the rockburst evaluation index，and MICE_RF algorithm was used to interpolate the missing data of rockburst data set.Then，a new combined weighting method was proposed，which is the improved analytic hierarchy process（IAHP）-weight determination method based on index correlation（CRITIC），and the weight of each index was comprehensively calculated by using the concept of weight vector distance. Finally，the ET algorithm was used to construct the rockburst prediction model after interpolation，weighting and normalization.Using the existing engineering example data at home and abroad，20 random sampling tests were carried out，and compared with other models to verify the superiority of this model in rockburst grade prediction.In this study，the interpolation effect based on MICE_RF missing value，the combined weighting effect of IAHP-CRITIC index and the comparison of the prediction effects of different models were analyzed and verified respectively.So，the ET rockburst prediction model based on MICE_RF and improved combined weighting was applied and the result of accuracy，precision，recall and RMSE were 93.10%，94.17%，93.44% and 0.2626.The results show that the MICE_RF missing data interpolation method not only increases the available rockburst data set，but also can effectively improve the prediction accuracy of three levels of no rockburst，intermediate rockburst and strong rockburst，and the average prediction accuracy of the complete data set has also been significantly improved.The improved AHP-CRITIC method has more advantages than the previous one，and the ET algorithm is significantly better than other comparison models in the results of four comparison indexes，that is，IAHP-CRITIC-ET model based on MICE_RF can significantly improve the prediction accuracy of rockburst grade，and the prediction results are more stable，which can provide effective guidance for similar projects.

山东黄金集团有限公司

http://www.goldsci.ac.cn/article/2022/1005-2518/1005-2518-2022-30-3-392.shtml