Predictive Model for FPSO Topsides Weight: A Data Science Case Applied to Surface Facilities

Abstract The estimation of topside weights for FPSOs in the early phases of projects is a key factor for the evaluation of their economic viability. Developing a method that provides this estimation in the initial phases is limited by the availability of the information. Also, the use of many input variables can lead to an overfitting problem. This work compares four regression models developed from a database containing information from 33 Petrobras' FPSO projects and selects the one that performs best based on two regression evaluation metrics: R2 and mean absolute percentage error (MAPE). The Random Forest Regression model achieved an R2 of 0,8869 and a MAPE of 0,1213. Finally, a web application is based on the model is presented, allowing end users to make their own estimations.