气候变化领域集成服务门户(CCPortal): Reduced-order models for the greenhouse gas leakage prediction from depleted hydrocarbon reservoirs using machine learning methods

CCPortal

DOI	10.1016/j.ijggc.2024.104072
	Reduced-order models for the greenhouse gas leakage prediction from depleted hydrocarbon reservoirs using machine learning methods
	Liu, Lei; Mehana, Mohamed; Chen, Bailian; Prodanovic, Masa; Pyrcz, Michael J.; Pawar, Rajesh
发表日期	2024
ISSN	1750-5836
EISSN	1878-0148
起始页码	132
卷号	132
英文摘要	Geologic storage of carbon dioxide (CO2) is one of the potential technological options to mitigate human -induced climate change. Depleted hydrocarbon reservoirs are promising candidates for storing CO2. However, these reservoirs contain residual hydrocarbons that may migrate beyond primary reservoirs during CO2 storage operations. Therefore, it is imperative to quantify the leakage risks of residual hydrocarbons (primarily methane) and CO2 to shallow aquifers. We focus on developing models that quantify leakage risks from wellbores that provide potential leakage pathways. To offset the computational intensity of high-fidelity reservoir simulations, we develop a Reduced -order model (ROM) enabling fast and accurate CO2 and hydrocarbon leakage predictions through wellbores. The ROM is generated using a dataset of 1000 high-fidelity, compositional reservoir simulations of CO2 injection into generic hydrocarbon reservoirs. The input parameters include reservoir depth, permeability, net to the gross ratio (NTG), reservoir pressure multiplier, wellbore permeability, average water saturation, oil compositions, and time. We analyze the performance of various ROM development techniques, including multivariate adaptive regression splines, gradient boosting, and neural networks. While all ROM development techniques yield excellent agreement with R2 higher than 0.95, the neural network model performs best compared to the other two methods. Furthermore, we explore developing a ROM as a collection of multiple sub -ROMs to improve accuracy across a wide range of predictions. We observe that using a set of sub -ROMs outperforms the prediction accuracy of a single ROM. Our work enables fast and reliable risk -assessment tools for CO2 geologic storage in depleted hydrocarbon fields.
英文关键词	Greenhouse gas; CO2 andCH4 leakage; Machine learning
语种	英语
WOS研究方向	Science & Technology - Other Topics ; Energy & Fuels ; Engineering
WOS类目	Green & Sustainable Science & Technology ; Energy & Fuels ; Engineering, Environmental ; Engineering, Chemical
WOS记录号	WOS:001162442200001
来源期刊	INTERNATIONAL JOURNAL OF GREENHOUSE GAS CONTROL
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/299604
作者单位	University of Texas System; University of Texas Austin; United States Department of Energy (DOE); Los Alamos National Laboratory; University of Texas System; University of Texas Austin
推荐引用方式 GB/T 7714	Liu, Lei,Mehana, Mohamed,Chen, Bailian,et al. Reduced-order models for the greenhouse gas leakage prediction from depleted hydrocarbon reservoirs using machine learning methods[J],2024,132.
APA	Liu, Lei,Mehana, Mohamed,Chen, Bailian,Prodanovic, Masa,Pyrcz, Michael J.,&Pawar, Rajesh.(2024).Reduced-order models for the greenhouse gas leakage prediction from depleted hydrocarbon reservoirs using machine learning methods.INTERNATIONAL JOURNAL OF GREENHOUSE GAS CONTROL,132.
MLA	Liu, Lei,et al."Reduced-order models for the greenhouse gas leakage prediction from depleted hydrocarbon reservoirs using machine learning methods".INTERNATIONAL JOURNAL OF GREENHOUSE GAS CONTROL 132(2024).