A DFN-based method for fast prediction of ships' added resistance in heading waves. (1st February 2022)

• Record Type: Journal Article
• Authors: Duan, Wenyang; Yang, Ke; Huang, Limin; Jing, Yu; Ma, Shan
• Is Part Of: Ocean engineering. Volume 245(2022)
• Publication Date: 2022-02-01
• Subjects: Added resistance -- Fast practical prediction -- Deep feedforward neural networks -- Heading waves; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2021.110484
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: The added resistance of ships in waves has been of great concern in recent years. In some engineering scenarios, rapid prediction of ships' added resistance is required. In this study, given the complexity of developing semi-empirical formulas and the small learning capacity of single hidden layer artificial neural networks, a method based on deep feedforward neural networks (DFNs) for fast prediction of ships' added resistance in heading waves is proposed. The proposed DFN-based method takes into account the particularity of added resistance prediction and makes innovations and optimizations in input vector parameters, the design of the input layer, the number of hidden layers, and the activation function of the output layer. A DFN-based model based on the proposed method, called DFN-AW, is constructed, achieving satisfactory results. Furthermore, the prediction accuracy of the DFN-AW model is better than generic DFN models, which proves the feasibility and advantage of the proposed method. Finally, the generalization ability of the DFN-AW model on new ships and speeds is investigated. Highlights: A method based on DFNs for fast prediction of ships' added resistance in heading waves is proposed. The DFN-based method avoids the process of formula design, and model parameters are automatically optimized in training. An input layer for added resistance feature is designed to standardize input vectors of the added resistance prediction. A power function-based nonlinear activation function is introduced into the output layer to make data distribute uniformly. The generalization ability of the DFN-based model on new ships or new speeds is explored and compared.