Development of neural network model based on attention mechanism applied to the prediction of ship damaged stability

Abstract

As a key indicator in ship design, many major incidents of ship sinking are related to the ship's damaged stability. The process of calculating the damaged stability becomes more and more complex and time-consuming on account of more and more stringent specification standards. A two-stage design step is used in this article to realize the calculation of ship’s damaged stability under various watertight bulkhead fast. Firstly, a multi-layer feed-forward neural network model was designed for the predictive regression of a ship's damaged stability using the location of the watertight bulkhead as a variable. Secondly, the relationship between each watertight bulkhead variant and the damaged stability A-value is analyzed. After that, with hydrostatic curve calculation based on the inlet simulation and the interaction between watertight bulkheads considered, a multilayer feed-forward neural network model based on the attention mechanism is designed, which could predict the regression of the damaged stability A-value and analyze bulkhead weights. Finally, the validity of the model was verified by the data, in which the mean value of the prediction error MAE (mean absolute error) was at 2.67×10^-4 and the computation time was greatly reduced.