Neural network prediction model for ultimate collision force in ship-bridge collision

Abstract

The numerical prediction of ultimate collision force in ship-bridge collision accident has a great influence on the safety of pier and bridge superstructure. However, ship-bridge collision problem is a complex nonlinear dynamic problem, and the linear empirical formula in the existing domestic and foreign standards cannot reflect the numerical value of collision force. In this paper, based on the data of established finite element model, the influence factors of maximum collision force are analyzed, four factors such as the time at which the maximum collision force occurs in the collision force curve, ship speed, hull quality and a coefficient which is related to the elastic deformation coefficient of the specimen of collision block (hull) and bridge pier are selected as the input of networks. BP and RBF neural network prediction models are established to predict the ultimate collision force of ship bridge collision. Combined with the prediction results, poor prediction accuracy of BP neural network is believed to be due to the over-fitting problem. Therefore GA-BP and PSO-BP algorithms are introduced to optimize the original BP neural network, of which the relative errors are 2.38% and 2.24% respectively and prediction accuracy is higher. The results show that it is feasible to predict the ultimate collision force of ship-bridge collision by using neural networks. Especially for the over-fitting problem of BP prediction model, the prediction accuracy of PSO-BP network after solving the over-fitting problem is convincing. The research in this paper successfully proves that the prediction of ultimate collision force by neural network in ship-bridge collision field is feasible with high precision accuracy, which provides scientific guidance and reference for the engineering safety of piers and other substructures in bridge design and construction.