Data-driven strategy for contact angle prediction in underground hydrogen storage using machine learning

In response to the surging global demand for clean energy solutions and sustainability, hydrogen is increasingly recognized as a key player in the transition towards a low-carbon future, necessitating efficient storage and transportation methods. The utilization of natural geological formations for underground storage solutions is gaining prominence, ensuring continuous energy supply and enhancing safety measures. However, this approach presents challenges in understanding gas-rock interactions. To bridge the gap, this study proposes a data-driven strategy for contact angle prediction using machine learning techniques. The research leverages a comprehensive dataset compiled from diverse literature sources, comprising 1045 rows and over 5200 data points. Input features such as pressure, injection rate, temperature, salinity, rock type, and substrate were incorporated. Various artificial intelligence algorithms, including Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Feedforward Deep Neural Network (FNN) and Recurrent Deep Neural Network (RNN), were employed to predict contact angle, with the FNN algorithm demonstrating superior performance accuracy compared to others. The strengths of the FNN algorithm lie in its ability to model nonlinear relationships, scalability to large datasets, robustness to noisy inputs, generalization to unseen data, parallelizable training processes, and architectural flexibility. Results show that the FNN algorithm demonstrates higher accuracy (RMSE = 0.9640) than other algorithms (RMSE RNN = 1.7452, RMSE SVM = 1.8228, RMSE KNN = 1.0582), indicating its efficacy in predicting the contact angle testing subset within the context of underground hydrogen storage. The findings of this research highlight a low-cost and reliable approach with high accuracy for estimating contact angle of water–hydrogen–rock system. This technique also helps determine the contribution and influence of independent factors, aiding in the interpretation of absorption tendencies and the ease of hydrogen gas flow through the porous rock space during underground hydrogen storage. • Hydrogen emerges as a leading clean energy solution, vital for a low-carbon future. • Underground storage methods ensure continuous energy supply and safety. • Machine learning tackles challenges in optimizing storage efficiency and reducing environmental risks. • Data-driven approach predicts gas-rock interactions for underground hydrogen storage. • FNN algorithm excels in accuracy, scalability, and robustness for contact angle prediction.