摘要
为了解决现有咸水中CO_(2)溶解度预测方法存在过程复杂、适用条件较窄和预测精度较低等问题,基于机器学习技术,提出了“权重重构+二次训练+深度优化”的建模方法,构建了一种基于LightGBM的咸水中CO_(2)溶解度预测模型。首先根据中国主要盆地咸水层的基本情况,广泛调研咸水中CO_(2)溶解度实验数据;然后综合TPE算法和五折交叉验证法,初步建立了基于LightGBM的咸水中CO_(2)溶解度预测模型;进而使用网格搜索法,针对决策树结构和袋外抽样频率,对预测模型进行了深度优化;最后采用不同指标评估了其综合性,并基于新建立的预测模型分析了CO_(2)溶解度的变化规律。研究结果表明:①新的CO_(2)溶解度预测模型的预测精度和可靠性高、泛化能力强,模型均方误差为0.00089(mol/kg)^(2),平均绝对百分比误差为3.78%,决定系数为0.994,综合性能明显优于常用的Duan&Sun、KRR、RBFNN-BAC和SVM等模型;②咸水中CO_(2)溶解度受温度的影响最大,压力次之,离子浓度最小,并且CO_(2)溶解度随温度的变化规律在20 MPa时出现反转。结论认为,该项研究成果可以为咸水层CO_(2)溶解封存潜力评估和封存场所筛选提供依据。
To address the problems in the existing methods for predicting CO_(2) solubility in brine such as complex process,narrow applicability,and low accuracy,a modeling method based on machine learning technology termed“weight reconstruction+secondary training+deep optimization”was proposed to construct a LightGBM-based prediction model.Firstly,experimental data of CO_(2) solubility in brine were extensively investigated according to the general situations of saline aquifers of major basins in China.Secondly,a LightGBM-based prediction model was preliminarily established using the TPE algorithm and five-fold cross validation.In addition,the model was deeply optimized with respect to the decision tree structure and out-of-bag sampling frequency using grid search method.Finally,the new model was employed to analyze the variation laws of CO_(2) solubility,after its comprehensive performance was evaluated through various indicators.The results show that the newly established CO_(2)solubility prediction model has high prediction accuracy and reliability,strong generalization ability,with a mean square error of 0.00089(mol/kg)^(2),a mean absolute percentage error of 3.78%,and a determination coefficient of 0.994,outperforming conventional models such as Duan&Sun,KRR,RBFNN-BAC and SVM.The CO_(2) solubility in brine is most affected by temperature,followed by pressure,and least by ion concentration.Moreover,the variation of CO_(2)solubility with temperature reverses at the pressure of 20 MPa.It is concluded that the research results can provide a basis for evaluating the potential of CO_(2) sequestration in saline aquifers and selecting suitable sequestration sites.
作者
张卫
戚会清
陈刚
赵安琪
胡世莱
ZHANG Wei;QI Huiqing;CHEN Gang;ZHAO Anqi;HU Shilai(School of Petroleum and Natural Gas Engineering,Chongqing University of Science and Technology,Chongqing 401331,China;Chongqing Key Laboratory of Complex Oil and Gas Field Exploration and Development,Chongqing 401331,China;School of Electronic and Electrical Engineering,Nanyang Technological University,Singapore 639798,Singapore)
出处
《天然气勘探与开发》
2025年第4期116-125,共10页
Natural Gas Exploration and Development
基金
重庆市教委科学技术研究项目“储气库多周期强注采下气—水两相渗流规律及近井区域盐析—运移—堵塞机制”(编号:KJQN202401501)。
