In order to solve the problem of poor stability and weighing accuracy of the load cell under ship load condition, the effect of ship swaying on the output of four-arm bridge resistive load cell was studied. The mass data of single fish at transverse tilt angle (0°~22.5°) and longitudinal tilt angle (0°~10°) were obtained, and analyzed by multivariate regression analysis on the output mass of the load cell and the tilt angle. A multi-layer BP neural network structure was constructed with the architecture of a 4-layer forward feedback [3-8-1-1] BP neural network, and the BP neural network was used to make predictions and adjust the parameters with a training model so that it could accurately predict the quality of the fish, and the algorithmic compensation method was studied. The results showed that the multiple regression coefficients estimated in the optimal linear equations reached the significant level (P<0.01) for different raw material qualities, indicating that the established linear regression equations had high reliability and good linearity, and the results of regression analysis and finite element analysis were consistent. The BP neural network model was used to construct the compensation method of shipboard instantaneous weighing data, and the actual mass value was predicted by the ship's transverse and longitudinal tilt angle, and the BP model showed good validity, high accuracy and good generalization ability for the change of weighing data under single action of transverse tilt and longitudinal tilt, as well as under the composite action, and the error rate of the BP model was reduced to 0.092% after compensation, which is very close to that of the mass value under the horizontal state, with a low error rate.The results of this research can provide a reference for the weighing of aquatic products under shipboard conditions.
