摘要
莺歌海盆地是南海新生代沉积盆地,具有显著的高温高压环境,地热资源潜力巨大。然而,由于海洋地区大地热流测量点分布有限,直接插值方法难以准确反映热流与盆地构造的耦合关系。为此,系统收集了980个训练集、244个测试集的大地热流测点数据及相应地质与地球物理数据,采用多源地质和地球物理数据融合的机器学习方法,利用线性回归、Lasso回归、随机森林、梯度提升和XGBoost等5种机器学习模型,刻画了该区地热热流特征,最后进行了热流预测性能对比,并对模型性能的影响因素进行了分析。研究结果表明:①XGBoost模型在测试集上表现出最高的R2值(0.72)、最低的均方根误差(16.62)和平均绝对误差(12.22),且训练集与测试集性能差异较小,展现出优异的泛化能力和预测稳定性;②基于XGBoost模型的预测结果显示,盆地热流分布存在明显的空间差异性,与区域构造格局密切相关,莺西斜坡为热流高值区,东北部莺东斜坡和西南缘则为低值区;③特征选择对模型性能具有较大影响,通过综合评估特征重要性并纳入局部特殊构造信息,可大幅提高模型预测精度,降低不确定性。结论认为,基于多源地质和地球物理参数融合的大地热流预测不仅填补了该盆地大地热流数据空白,完善了其热流分布特征,有助于更准确地刻画盆地热流分布特征及其与构造的耦合关系,同时也为其他海洋沉积盆地的地热资源评估提供了技术参考。
As a Cenozoic sedimentary basin in the South China Sea,the Yinggehai Basin exhibits a notably high-temperature and high-pressure environment and possesses significant geothermal resource potential.Due to the limited distribution of terrestrial heat flow measurement points in marine areas,however,the direct interpolation method can hardly reflect the coupling relationship between heat flow and basin structure accurately.To deal with this situation,this paper systematically collects the terrestrial heat flow measurement data of 980 training sets and 244 testing sets and their corresponding geological and geophysical data.Then,the characteristics of geothermal heat flow in this area are characterized by means of the multi-source geological and geophysical data integrated machine learning method,which adopts five machine learning models,including linear regression,Lasso regression,random forest,gradient boosting,and XGBoost.Finally,their performances in heat flow prediction are compared,and the factors influencing model performance are analyzed.The following results are obtained.First,the XGBoost model presents the highest R2 value(0.72)and the lowest Root Mean Square Error(RMSE:16.62)and Mean Absolute Error(MAE:12.22)on the testing set.Moreover,the training and testing sets are less different in the performance,showing excellent generalization ability and prediction stability.Second,the prediction results based on the XGBoost model reveal significant spatial difference in the heat flow distribution within the Yinggehai Basin,which is closely related to the regional tectonic pattern.The Yingxi Slope is a high heat flow zone,while the northeastern Yingdong Slope and the southwestern margin are low-value zones.Third,feature selection has a significant impact on model performance.Comprehensively evaluating feature importance and incorporating the information of local special structures can substantially improve prediction accuracy and reduce uncertainty.In conclusion,the terrestrial heat flow prediction based on integrated multisource geological and geophysical parameters not only fills the gap in the terrestrial heat flow data for the Yinggehai Basin and refines its heat flow distribution characteristics,enabling a more accurate characterization of the basin's heat flow distribution and its coupling relationship with tectonics,but also provides a technical reference for geothermal resource assessment in other marine sedimentary basins.
作者
郑华安
宋荣彩
宋吉锋
陈海雯
董奔
张佳豪
张严
郑峰
ZHENG Hua'an;SONG Rongcai;SONG Jifeng;CHEN Haiwen;DONG Ben;ZHANG Jiahao;ZHANG Yan;ZHENG Feng(CNOOC China Limited Hainan Company,Haikou,Hainan 570312,China;State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation//Chengdu University of Technology,Chengdu,Sichuan 610051,China;College of Energy,Chengdu University of Technology,Chengdu,Sichuan 610051,China;College of Earth and Planetary Sciences,Chengdu University of Technology,Chengdu,Sichuan 610051,China;Sichuan Institute of Comprehensive Geological Survey,Chengdu,Sichuan 610081,China)
出处
《天然气工业》
北大核心
2025年第8期170-180,共11页
Natural Gas Industry
基金
国家自然科学企业创新发展联合基金项目“南海北部深水区深层温压场差异性演化对优质储层的控制机理”(编号:U24B2016)。
作者简介
郑华安,1979年生,高级工程师,硕士,主要从事油气田开发生产工作。地址:(570312)海南省海口市秀英区仲韶街88号。ORCID:0009-0008-9249-5704。E-mail:zhengha2@cnooc.com.cn;通信作者:宋荣彩,1975年生,教授,博士,主要从事地热地质学、油气储层地质等研究工作。地址:(610051)四川省成都市成华区二仙桥东三路1号。E-mail:songrongcai06@cdut.edu.cn。
