基于SMOTEENN-CGAN-Stacking的岩爆烈度等级预测研究  

ROCKBURST INTENSITY LEVEL PREDICTION BASED ON SMOTEENN-CGAN-STAKCING

作　　者：高梅 张成良[1] 张华超 吴泽鑫 GAO Mei;ZHANG Chengliang;ZHANG Huachao;WU Zexin(Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China)

机构地区：[1]昆明理工大学,国土资源工程学院,昆明650093

出　　处：《工程地质学报》2024年第6期2264-2276,共13页Journal of Engineering Geology

基　　金：国家自然科学基金项目(资助号:51934003)。

摘　　要：随着地下工程的不断发展和扩大规模,岩爆灾害在施工过程中频繁发生,对工程及施工人员生命造成了严重威胁。因此,岩爆烈度等级预测成为防范岩爆灾害的重要的研究方向。本文选取围岩最大切应力σ_(θ)、单轴抗压强度σ_(c)、单轴抗拉强度σ_(t)和弹性能量指数W_(et)作为预测模型的4个特征值,提出了一种基于SMOTEENN-CGAN数据处理的Stacking集成算法的组合模型,用于岩爆烈度等级的预测。在该模型中,首先使用SMOTEENN和CGAN算法以过采用、欠采样、对抗生成的方法处理原始数据;随后采用10种经典算法验证SMOTEENN-CGAN的有效性;最后以Stacking集成算法构建出4组含不同基模型和元模型的岩爆烈度等级预测模型。结果表明:(1)SMOTEENN-CGAN能用于处理多分类问题,新生成的岩爆数据符合原始分布特征,预处理后的数据特征值离散程度,异常点明显减少;(2)数据经过预处理后,10种经典算法的性能得到不同程度的提升,各算法的平均准确率提高了1.87%~7.75%不等;其中MLP与NP提高较多,分别为7.75%与7.43%。(3)不同的基模型与元模型的搭配会影响Stacking的性能,在组合(4)中,基模型为XGBoost+LGBM+ETC时,元模型中的Adaboost最高预测准确率为96.12%。通过工程实例验证Stacking岩爆烈度等级预测模型的可靠性时,预测最高准确率可达92.3%。本文模型为岩爆烈度预测提供了一种有效可行的机器学习预测方法。Rockburst disasters constantly occur in the construction of mines,tunnels(caves),foundation pits,and other underground projects at home and abroad,and their frequency and intensity increase with the extension of projects into deeper areas.Therefore,predicting the intensity of rockburst levels has become an important research direction.By selecting maximum shear stress(σ_(θ)),uniaxial compressive strength(σ_(c)),uniaxial tensile strength(σ_(t)),and elastic energy index(W_(et))as four characteristic values of the prediction model,this paper presents a combined model based on the SMOTEENN-CGAN data processing integrated algorithm that is used to predict the rockburst intensity level.In this model,SMOTEENN and CGAN algorithms are used to process the original data through the methods of over-sampling,under-sampling,and augmented generation.Then,ten classical algorithms are used to verify the validity of SMOTEENN-CGAN.Finally,four groups of rockburst intensity level prediction models with different base models and meta-models are constructed with the Stacking integrated algorithm.The results indicate that:(1)SMOTEENN-CGAN can be used to deal with multiple classification problems,and the newly generated rockburst data is consistent with the original distribution characteristics,and the dispersion degree of pre-processed data and abnormal points are significantly reduced;(2)After data preprocessing,the performance of ten classical algorithms has been improved to varying degrees,and the average accuracy of each algorithm has increased by 1.87%to 7.75%;MLP and NP increased by 7.75%and 7.43%,respectively.(3)Different combinations of base models and meta-models affect the performance of Stacking.In combination④,when the base model is XGBoost+LGBM+ETC,the highest prediction accuracy rate of Adaboost in the meta-model is 96.12%.When the reliability of the rockburst intensity level prediction model is verified by engineering examples,the maximum prediction accuracy can reach 92.3%.The model in this paper provides an effec

关 键 词：不平衡数据集 CGAN SMOTEENN STACKING 岩爆烈度等级预测 

分 类 号：TU457[建筑科学—岩土工程]