AI-driven surrogate modeling and design optimization of the nitrogen single-expander LNG process for energy efficiency improvement

Abstract

Natural gas liquefaction is an energy-intensive operation in the LNG value chain. The N₂ single-expander process is widely used in small- to mid-scale plants because of its simplicity, safety, and operational flexibility, making it ideal for modular and offshore applications. However, its efficiency remains significantly lower than mixed-refrigerant cycles. While recent literature explores AI-assisted optimization for LNG processes, most approaches rely on computationally intensive, iterative simulator-in-the-loop frameworks. To address this gap, this study develops a fully decoupled, AI-based surrogate framework that instantaneously predicts optimal operating variables (e.g., pressures, temperatures, and refrigerant flow rates). By strictly enforcing thermodynamic feasibility constraints, such as the MITA, natively within the model, this approach eliminates the need for repeated runtime simulations. The framework achieves remarkable predictive accuracy, reducing the MITA prediction error to 8.21% and maintaining the specific energy consumption error at approximately 6.7% when trained on 10,000 samples. It quantitatively outperforms a benchmarked ANN, which yielded an 11.13% MITA error under identical conditions. Furthermore, the model successfully identifies optimal operating windows capable of reducing overall SEC by up to 15% compared to unoptimized baseline operations. Ultimately, this work transforms simulation-heavy LNG process design into a rapid, accurate, and scalable decision-support tool. By enhancing the efficiency of the N₂ single-expander process, this data-driven framework accelerates LNG digitalization, boosts industrial competitiveness, and directly contributes to global decarbonization targets.