Statistical proxy modeling for life cycle assessment and energetic analysis

Bottom-up life-cycle assessment (LCA) based on engineering-based models is emerging as a way to model the energy consumption and greenhouse gas emissions from important segments of the energy sector. However, a major challenge that has hindered further application of this approach is data and computationally intensive thermodynamic modeling. In order to address this issue, we introduce a general data-driven framework to develop statistical reduced-order models (henceforth “proxy models”) from advanced thermodynamic or engineering simulations that can be utilized for bottom-up LCA and/or energetic assessment purposes of any energy system. To demonstrate the performance of the proposed framework, we simulate four important oil and gas process units with a commercial process simulation package. Using a combination of deterministic and random sampling strategy, >25,000 simulations are performed and quadratic proxy models are trained on the results to predict the energy consumption and product compositions across a wide ranges of independent variables. The simple proxy models have excellent predictive accuracy (R2 > 0.95 in most cases, R2 = 1 for pump power consumption prediction), solve nearly instantaneously, and require fewer input parameters (less than 10, varies based on the process unit). We also examine and prove the proposed methodology stability across training and validation datasets via 1,000 independent data splitting runs. We lastly implement the oil and gas proxy models in a LCA simulator, compare the results with textbook correlations, and demonstrate the improved flexibility.