A Machine Learning Framework for Data Driven Platform Electrical Systems Design and Simulation

Abstract

Abstract FPSO platforms have traditionally employed conservative designs for their electrical systems, resulting in lower load conditions than originally anticipated during the design phase. This work proposes a machine learning framework for data driven FPSO electrical systems design and simulation, with several machine learning and data analysis routines being implemented. The user experience design was done with the objective of the tool being easy to understand and use by engineering teams responsible for management level decisions. To achieve this an iterative development process was employed using target users’ feedback, with the final automatic regression functionality achieving an average mean absolute error of 0.87% per platform on validation data. Some features, such as the clustering functionality were specifically requested during the development, while others, such as the automated data preprocessing pipeline, were designed to minimize the need for user interference while maintaining a high-quality dataset for the user to work with. Developed in the background of a partnership between a research team at UFF and Petrobras, the framework was found appropriate and is currently at an early adoption phase by relevant teams at Petrobras.