| 雷磊,付雷.多源数据融合的二氧化碳地质封存潜力预测方法研究[J].矿产勘查,2026,17(S1):331-340 |
| 多源数据融合的二氧化碳地质封存潜力预测方法研究 |
| Research on prediction methods for CO2 geological storage potential based on multi-source data fusion |
| 投稿时间：2026-04-22 修订日期：2026-05-30 |
| DOI：10.20008/j.kckc.2026S1038 |
| 中文关键词: 二氧化碳地质封存 多源数据融合 随机森林插补 机器学习 潜力预测 |
| 英文关键词: CO₂ geological storage multi-source data fusion random forest imputation machine learning potential prediction |
| 基金项目:本文受国家自然科学基金专项项目（42141009）资助。 |
|
| 摘要点击次数: 4 |
| 全文下载次数: 1 |
| 中文摘要: |
| 为解决二氧化碳地质封存（CCUS）潜力预测关键指标获取难、预测误差大的问题，本文提出了基于“类比+数值模拟”双样本库的多源数据融合预测方法。采用随机森林迭代插补完善缺失数据，通过皮尔逊相关系数法筛选关键预测指标，采用多元线性回归等五种机器学习算法并行建模，最终建立CCUS潜力与丰度定量预测关系式。结果表明：非线性回归模型为两类样本库的最优预测模型，模拟样本库潜力、丰度拟合决定系数（R²）分别达 0.9211、0.9179，预测准确率约70%；类比样本库经插补后模型性能提升约40%，拟合R²达0.7468，预测准确率超45%。本研究构建了从数据完善到模型应用的一体化闭环预测体系，为CCUS场地筛选和资源评价提供了可靠的技术工具。 |
| 英文摘要: |
| To solve the problems of difficult acquisition of key indicators and large prediction errors in traditional methods for CO₂ geological storage （CCUS） potential prediction， this paper proposes a multi-source data fusion prediction method based on the "analogy + numerical simulation" dual sample library. Random forest iterative imputation is used to supplement missing data， key prediction indicators are selected via the Pearson correlation coefficient method， and five machine learning algorithms， such as multiple linear regression and nonlinear regression， are adopted for parallel modeling. The optimal model is determined with the maximum coefficient of determination （R²） as the core goal， and quantitative prediction formulas for CCUS potential and abundance are finally established. The results show that the nonlinear regression model performs best for both sample libraries. For the simulation sample library， the R² values of potential and abundance fitting are 0.9211 and 0.9179， respectively， with a prediction accuracy of about 70%. For the analogy sample library， the model performance is improved by approximately 40% after data imputation， with a fitting R² of 0.7468 and a prediction accuracy of over 45%. This study constructs an integrated closed-loop prediction system from data completion to model application， establishes and verifies four types of prediction formulas， and provides a reliable technical tool for CCUS site screening and resource assessment. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |