Prediction of main particulars of container ships using artificial intelligence algorithms
Abstract
The value of the main ship particulars are key values to determine during initial design of a vessel, but they can be complex to determine, as they depend on a large number of factors. The presented research attempts to model the main particulars: length between perpendiculars (LPP), length overall (LOA), modulated breadth (B), depth (D), draught (T), gross tonnage (GT), net tonnage (NT), deadweight (DWT), and engine power, using only key request factors: number of twenty-foot equivalent units (TEU) the vessel is expected to carry, and the expected speed of the ship (V). As this is a complex task, artificial intelligence (AI) techniques are applied to the dataset consisting of 250 container ships. Two modeling techniques are used: multilayer perceptron (MLP) and gradient boosted trees (GBT). The model hyperparameters are trained using a grid search procedure and evaluated using mean absolute percentage error (MAP E) and coefficient of determination (R2) in a 5-fold cross-validation scheme. The obtained results show that a quality model can be achieved using both techniques, except in the case of engine power for which a high-quality model has not been regressed. Models presented here can have practical application in the determination of the ship's main particulars at the preliminary design stage.