A Holistic Approach to the Development of Digital Twin Advisor for Upstream Production Process

Abstract

From upstream production process/operation perspectives, adoption of digital twin linked with real-time data for operational monitoring and advisory can assist in early determination of abnormalities that can result in potential failure events, based on deviation from the operating setpoints and limits. In addition, prediction capability of the digital twin model can assist in prevention of unplanned production deferment and proper maintenance planning. The digital twin model can also recommend optimum operating points based on changes in the system as operational guidance to further improve and optimize the system performance. This paper outlines the holistic approach taken for the development of digital twin-based advisory system for upstream process monitoring and optimization. Based on a case study, a step-by-step approach for the development of a digital twin model for fuel gas system is being described. This technology combines various aspect of process and operation models. The Integrated Operating Limit Model acts as the first layer of digital twin model, which provides the operators with a quick assessment based on the visualization or representation of the real time operating points versus the operating set points and limits defined in the system. Process simulation model predicts other operating parameters not covered by the existing instrumentations, serving as virtual analysers, providing near real-time equipment and unit performance. Data driven model can be used for prediction, optimization and anomaly detection, and based on the prediction model development, comparison between linear regression (LR) versus decision forest-based model showed LR to be more reliable approach based on the validation against the actual operation data especially during upset condition (R2 of 0.9 vs 0.27). A graphical user interface was also developed to translate and summarize the model outputs into a clear visualization to be used by the operation team as advisory and surveillance tool. Combination of these models’ outputs would allow operator to essentially screen for potential process upset and apply further optimization based on the recommendation provided.