Abstract: Objectives By addressing the issues of insufficient coverage, model system deviation and strong noise in traditional data-driven methods for establishing a Wave-Added Resistance(WAR) prediction model for actual ship to enhance the forecasting reliability of merchant ships and the minimum main-engine power, a WAR physics-data dual-driven model integrating the physical mechanisms and data-driven approaches is constructed by the usage of a small amount of model tests and numerical simulation data for amendment.Methods The results of model test and numerical calculation based on the potential flow theory for a certain ocean-going bulk carrier are systematically analyzed to construct a physical data set for WAR on the basis of experiment and theoretical models. The WAR database is separated by proposing a strategy of wind-wave added resistance iterated with propulsion efficiency and constructed from the operational data of actual ship. Relying on the physical data sets of model experiments and numerical calculations, the separated WAR database of actual ship is amended by using interpolation and assimilation methods, while comparative and statistical analyses are carried out under different sea conditions. Results The results show that the forecast errors of WAR database amended by the experimental-numerical fusion of physical data sets is reduced with the accuracy improved. The dual-driven model by amendment of RBF and Lagrange interpolation possesses the highest accuracy at WWO sea conditions of level 3 and level 4, while statistical correlation <italic>r</italic> value between the predicted and true results reaches over 0.93 with <italic>MSE</italic> less than 1.5%, respectively. Conclusions It is concluded that the proposed physical-data dual-driven WAR prediction model for real ships by physical experiments and shipping big data reveals better prediction accuracy and engineering applicability compared to single data-driven models, which provides an important support for the estimation of minimum propulsion power and the verification of model interpretability.