This study investigates the primary factors influencing the power output of offshore floating photovoltaic (FPV) systems, with a focus on informing operational and maintenance strategies for renewable energy systems. Utilizing multidimensional field data collected from the Yellow Sea region of China, three tree-based regression models—Random Forest (RF), XGBoost, and LightGBM—were developed and assessed. Model interpretability was enhanced through SHapley Additive exPlanations (SHAP). Among the evaluated models, the RF model demonstrated superior predictive accuracy, achieving coefficients of determination (R2) of 0.980 and 0.946 for the training and testing datasets, respectively, and the lowest root mean square error (RMSE) and mean absolute error (MAE). SHAP analysis identified frontal irradiance as the most influential factor, contributing 59.2% to total feature importance, followed by sea surface albedo (23.5%), module backplane temperature (13.0%), and wind speed (4.2%). These findings provide a data-driven basis for optimizing the design and operational strategies of offshore FPV systems, highlighting the value of interpretable artificial intelligence in supporting renewable energy technology development and maintenance. The insights gained from this study can inform targeted maintenance practices, improve system reliability, and enhance the overall efficiency of offshore FPV installations.