In this paper, a multi-objective optimization design is carried out for the total drag of DTMB-5415 ship in calm water at the speeds of three Froude numbers. Firstly, 29 design variables are set up based on the Free Form Deformation method, and the design space are later reduced by Proper Orthogonal Decomposition method. Combined with the multi-objective genetic algorithm, the optimal hulls with better comprehensive resistance performance are obtained, and optimization effect is analyzed through further verification. Furthermore, based on Proper Orthogonal Decomposition method, the ship flow field (free-surface wave elevation and hull surface pressure) learning considering ship speed and shape parameters is constructed to realize the quick prediction of ship calm-water flow field in a large speed range.
For the hull form optimization problem, if the number of design variables determined by hull form deformation method is small, then in a certain design space, the hull shape changes may be small, so the improvement of hydrodynamic performance of the optimal hulls found in the low-dimensional design space is relatively limited; on the contrary, if there are many parameters controlling the changes of the hull shapes, the possible changes in shape can be more abundant within the range of design variables, i.e., more varied shape changes can be produced in the high-dimensional design space. Theoretically, it is possible to find the optimal hull shape with better performance.
The so-called "dimensionality reduction" means to reduce the data dimension as much as possible on the premise of containing all the information in a certain number of high-dimensional data. By reducing the dimensionality of the original design space composed of the hull form deformation parameters (design variables), it is possible to characterize any possible change of the hull shape in the original high-dimensional design space by using a small number of design variables without losing information as far as possible. In this way, the experimental design can be carried out in the dimensionality-reduced design space, and the number of sample points required for high-fidelity surrogate model construction, so the computational cost will be reduced to speed up the optimization efficiency.