In the quest to curb carbon emissions, scientists are turning to machine learning to optimize a critical process: geological sequestration of CO2. A recent study published in the journal *Nature Scientific Reports* by Kashif Liaqat and colleagues from the Department of Mechanical Engineering at Rice University presents a promising breakthrough in this area. The research focuses on predicting the interfacial tension (IFT) between CO2 and brine (salt water), a key factor in determining the storage capacity of CO2 in saline aquifers.
Geological sequestration is considered one of the most viable options for reducing CO2 emissions, but it faces challenges. Measuring IFT experimentally is time-consuming, labor-intensive, and expensive. Moreover, existing empirical correlations often fall short in accuracy. This is where machine learning (ML) comes into play. “We wanted to explore whether ML models could offer a more efficient and cost-effective alternative to traditional methods,” says Liaqat.
The study compares the performance of various ML models, including linear regression, support vector machines (SVM), decision tree regressors, random forest regressors, and multilayer perceptrons (MLP). The models were trained on a dataset covering a wide range of temperature, pressure, and salinity conditions. After optimizing each model through hyperparameter tuning, the researchers evaluated their performance using metrics like mean absolute error (MAE) and mean absolute percentage error (MAPE).
The results were impressive. The SVM and MLP models emerged as the top performers, with MAPE values of 0.97% and 0.99%, and MAE values of 0.39 mN/m and 0.40 mN/m, respectively. In contrast, the linear regression model had the worst performance, with a MAPE of 4.25% and an MAE of 1.7 mN/m. “The accuracy of these models is a significant improvement over previous studies,” Liaqat notes.
The research also revealed that pressure is the most influential parameter affecting IFT. This insight could guide future experiments and models, potentially leading to more accurate predictions and better designs for CO2 sequestration.
The implications for the energy sector are substantial. Accurate IFT predictions can enhance the efficiency and cost-effectiveness of CO2 storage, making geological sequestration a more attractive option for reducing carbon emissions. “Our findings demonstrate that even basic ML models, when properly tuned, can provide accurate predictions,” Liaqat explains. “This could revolutionize the way we approach CO2 sequestration, offering a faster, cheaper, and more reliable method.”
As the world grapples with the challenges of climate change, innovations like this one offer a glimmer of hope. By harnessing the power of machine learning, scientists are paving the way for more effective carbon capture and storage strategies, ultimately contributing to the global effort to achieve carbon neutrality. The study, published in *Nature Scientific Reports*, underscores the potential of ML in transforming the energy sector and accelerating the transition to a low-carbon future.