The invention discloses a ship shape design proxy auxiliary optimization method based on transfer learning, and belongs to the field of ship shape design. Firstly, samples represented by two different types of hull shape parameters are utilized, and a proxy model is constructed by implementing a two-stage TrAdaBoost.R2 transfer learning algorithm. The two parameter representations are based on a ship scale coefficient (used as a source domain) and a geometric parameter (used as a target domain), respectively. Secondly, a proxy model updating strategy is provided, and two CFD simulation positions with the current optimal estimation fitness value and the maximum uncertainty are added to a target domain at regular intervals so as to reconstruct a substitution model; and finally, an improved quantum particle swarm optimization (QPSO) algorithm combined with a particle migration scheme is adopted as an optimizer, so that the algorithm can better adapt to reconstruction of the substitution model.
基于迁移学习的船体外形设计代理辅助优化方法,属于船型设计领域。首先,利用两种不同类型的船体形状参数表示的样本,通过实施两阶段TrAdaBoost.R2迁移学习算法构建代理模型。这两种参数表示分别基于船舶尺度系数(用作源域)和几何参数(用作目标域)。其次,我们提出了一种代理模型更新策略,定期将当前最佳估计适应度值和最大不确定性的两个CFD模拟位置分别添加到目标域,以重构替代模型。最后,采用了结合粒子迁移方案的改进量子粒子群优化(QPSO)算法作为优化器,以使其更好地适应替代模型的重构。
Agent aided optimization method for ship shape design based on transfer learning
基于迁移学习的船体外形设计代理辅助优化方法
2024-05-10
Patent
Elektronische Ressource
Chinesisch
NTIS | 1971
|Computer-aided ship design at MarAd
Tema Archiv | 1975
|Computer-aided preliminary ship structural design
TIBKAT | 1981
|