Session: 15-07-01 Ship Manoeuvring, Resistance and Propulsion

Submission Number: 180134

Residual Resistance Prediction of Sailing Yachts Based on Optimized Random Forest Algorithm 

The residual resistance prediction of sailing yachts is critical to their performance, as it directly influences key indicators like sailing efficiency, energy consumption, and maneuverability. Traditional prediction methods, including empirical formulas and simplified numerical simulations, often struggle with accuracy and adaptability. They rely on linear assumptions or oversimplified physical models, which fail to address the nonlinear, multi-factor nature of sailing yacht residual resistance. To solve this problem, this paper establishes a Random Forest (RF) regression model to predict the residual resistance per unit weight of displacement of sailing yachts. The RF model is suitable for this task due to its anti-overfitting ability, tolerance to noise data, and capacity to capture complex interactions between input variables.

This study targets two common issues in model input data: non-normal distribution and multicollinearity. Non-normal distribution may bias model training, while multicollinearity can lead to redundant feature information and unstable predictions. To tackle these issues, a two-step preprocessing strategy is adopted. First, Box-Cox transformation is used to normalize the input data, adjusting skewed distributions to approximate a normal distribution and improving the model’s ability to learn correlations between features and targets. Second, Principal Component Analysis (PCA) is applied to eliminate multicollinearity among high-dimensional features. This step reduces data dimensionality while retaining main information of the original features, and avoids the “curse of dimensionality” that may harm prediction accuracy.

To further improve the RF model’s performance, five hyperparameter optimization algorithms are compared, including Random Search (RS), Bayesian Optimization (BO), Hyperband (HB), BOHB (a hybrid of BO and HB), and Genetic Algorithm (GA). Experimental results show that the RF model optimized by RS outperforms the other four algorithms. After Box-Cox transformation and PCA dimensionality reduction, the model’s Mean Squared Error (MSE) is reduced by 24.09%, which highlights the value of feature engineering in data-driven models. With the optimal hyperparameter combination from RS, the model’s Root Mean Squared Error (RMSE) is further reduced by 21.1%, and its coefficient of determination (R²) is higher than that of models optimized by BO, HB, BOHB, or GA.

Compared with existing models such as Support Vector Machine (SVM), and Neural Network (NN) that use the same sailing yacht residual resistance dataset, the RS-optimized RF model achieves an order-of-magnitude improvement in prediction accuracy. This study confirms that the RF algorithm, combined with targeted feature engineering and hyperparameter optimization, provides a reliable method for sailing yacht residual resistance prediction, and can serve as a reference for subsequent hull design optimization and marine engineering data-driven research.

Presenting Author: Yong Zhao Dalian Maritime University

Presenting Author Biography: Professor Yong Zhao is affiliated with the School of Naval Architecture and Ocean Engineering, Dalian Maritime University. His main research focus is on the integrated application of hydrodynamics, artificial intelligence, numerical ocean dynamics, and fluid mechanics in the field of deep-sea mining.
In his research work, he employs open-source Computational Fluid Dynamics (CFD) tools such as OpenFOAM, AI platforms including TensorFlow/PyTorch, and ocean dynamics simulators like FVCOM/Telemac to conduct relevant studies. Meanwhile, he leads a graduate research team, focusing on technological exploration and academic innovation in the aforementioned fields.
He has published a number of papers in Ocean Engineering and Physics of Fluids, and has led several scientific research projects.
Please refer to https://orcid.org/0000-0001-8216-9210 (ORCID profile).

Authors:

Yong Zhao Dalian Maritime University
Xuehui Zhao Dalian Maritime University