Objective The aim is to enhance energy consumption prediction accuracy in floating production storage and offloading (FPSO) process systems and prevent equipment power shortages or overpressure incidents.

Method Firstly, this study investigated an oil-gas processing system of a Brazilian offshore FPSO. A HYSYS process simulation model was developed. Secondly, a Bayesian-optimized XGBoost (BO-XGBoost) prediction model was constructed using simulation data, optimizing key parameters including decision tree quantity, maximum tree depth, minimum leaf node weight, and learning rate. Thirdly, Shapley additive explanations (SHAP) analysis was employed to quantitatively assess factors’ contributions and importance, and the overall influence of feature to model's performance in global range was obtained.

Result Firstly, compared with back propagation neural network (BP), random forest (RF) and standard XGBoost models, the coefficient of determination (R²) of the BO-XGBoost model increased by an average of 0.09, 0.06, and 0.03, respectively; the mean value of mean absolute percentage error (MAPE) of the improved model was reduced by 11 percentage points, 6 percentage points, and 4 percentage points, respectively. The analysis framework of the optimized model had higher fitting accuracy for energy consumption. Secondly, valve openings affecting fuel/reinjection gas distribution showed bidirectional effects, while higher crude throughput increased energy consumption of upper FPSO process module. Thirdly, variables including oil treatment system allocation valves opening, reinjection gas and deacidification acid gas valves opening, inlet pressures, and vapor recovery system compressor outlet pressures, dehydration system inlet pressures exhibited limited influence (narrow SHAP distribution ranges), suggesting field operational adjustments should consider actual conditions and fuel gas consumption.

Conclusion Identifying key energy consumption factors and establishing the BO-XGBoost model can achieve precise energy consumption prediction of process system, thereby effectively preventing power shortages/overpressure, ensuring safe FPSO operations, improving energy efficiency, and reducing operational costs.