(273e) A Novel Data-Driven Approach to Reducing the Environmental Footprint of an Integrated Gas-Oil Separation Network | AIChE

(273e) A Novel Data-Driven Approach to Reducing the Environmental Footprint of an Integrated Gas-Oil Separation Network


del Rio Chanona, E. A., Imperial College London
Shah, N., Imperial College London
As the pressure to meet net-zero targets builds, the energy industry, particularly the oil and gas sector, is investigating various opportunities spanning the supply chain. Accordingly, companies are considering the adoption of advanced analytic techniques to monitor operations performance and reduce their environmental footprint. However, achieving this objective while maintaining a solid financial performance is critical, thus, requiring innovated and integrated solutions.

This work has focused on the upstream business since it accounts for a substitutional percentage of profit margin and emissions. In large oil reservoirs, a cluster of producing wells supply a network of gas-oil separation facilities interconnected via swing pipelines. The primary function of the facilities is to separate the three phases and remove contaminants. Nonetheless, they play a vital role in eliminating flaring and greenhouse gas emissions by recovering associated gases.

For the past few years, scientists and practitioners have devoted several efforts to optimizing network operations. However, it was focused on achieving higher savings in operating expenditures (OPEX) while overlooking sustainability. Furthermore, previous contributions had several limitations in the proposed method—for example, they utilized a physics-based model with expensive computation time.

Due to these limitations, we have developed a novel data-driven method that addresses them. Our approach uses artificial neural networks in a mixed integer linear programming (MILP) framework to quantify the entire network's environmental footprint and operating expenditures, then optimize it by manipulating the feed allocation and equipment utilization. The surrogate models are trained using a high-fidelity simulation model to capture possible feed uncertainties. Furthermore, the proposed method has been tested using several production scenarios obtained from actual production rates and demonstrated outstanding capabilities. As a result, it has significantly reduced greenhouse gas emissions while maximizing profitability.