混合ABC-Stacking机器学习的钻孔数据地层三维隐式建模方法

打开文本图片集

[分类号］P62 [文献标志码］A

Abstract： Thre-dimensional stratigraphic modelscan directly and accurately reflect the geological structure characteristics of underground space，which is of great significance for underground space development and utilization.However，limited borehole geological exploration data makes constructing detailed thre-dimensional stratigraphic models dificult.This paper proposes ahybrid stacking machine learning strategybased on borehole data. Virtual borehole network datasetsare built from limited existing borehole data，whichare then used for threedimensional implicit stratigraphic modeling.First，theartificial beecolonyalgorithm isused tocreate anoptimized stacking ensemble learning model using several common machine learming algorithms.The stratigraphic distribution paterns of existing borehole data are learned， which are then used to construct virtual borehole clasification data. Then，a three-dimensional stratigraphic model isbuilt basedonthe implicitradial basis functionmodeling method. Finally，aquantitativeindicatorof geological profilecoincidenceisintroduced formodel evaluation.Case study resultsshow that the stacking ensemble learning model achieves an Fl score and accuracy of 88% and 89% 0 respectively，forthetestset，demonstrating higherclasificationaccuracycompared tosinglemachinelearmingmodels instratigraphic prediction.The approach effctively improves the precision of local stratigraphic classification， with an average coincidence of 78.38% between the constructed three-dimensional model profiles and actual geological survey sections，further confirming the efectiveness of the proposed method and providing insights for refined threedimensional modeling of underground structures.

Key words： three-dimensional stratigraphic modeling； stacking method； implicit modeling；machine learning;artificial bee colony

地层三维模型能直观透明地展示地下空间地层界面和三维结构，为城市地下空间开发利用、地质灾害调查以及场地土壤-地下水环境污染分析提供了重要的数据和技术支持。（剩余21688字）